1afc51aaaSDarrick J. Wong // SPDX-License-Identifier: GPL-2.0 2afc51aaaSDarrick J. Wong /* 3afc51aaaSDarrick J. Wong * Copyright (C) 2010 Red Hat, Inc. 4598ecfbaSChristoph Hellwig * Copyright (C) 2016-2019 Christoph Hellwig. 5afc51aaaSDarrick J. Wong */ 6afc51aaaSDarrick J. Wong #include <linux/module.h> 7afc51aaaSDarrick J. Wong #include <linux/compiler.h> 8afc51aaaSDarrick J. Wong #include <linux/fs.h> 9afc51aaaSDarrick J. Wong #include <linux/iomap.h> 10afc51aaaSDarrick J. Wong #include <linux/pagemap.h> 11afc51aaaSDarrick J. Wong #include <linux/uio.h> 12afc51aaaSDarrick J. Wong #include <linux/buffer_head.h> 13afc51aaaSDarrick J. Wong #include <linux/dax.h> 14afc51aaaSDarrick J. Wong #include <linux/writeback.h> 15598ecfbaSChristoph Hellwig #include <linux/list_sort.h> 16afc51aaaSDarrick J. Wong #include <linux/swap.h> 17afc51aaaSDarrick J. Wong #include <linux/bio.h> 18afc51aaaSDarrick J. Wong #include <linux/sched/signal.h> 19afc51aaaSDarrick J. Wong #include <linux/migrate.h> 209e91c572SChristoph Hellwig #include "trace.h" 21afc51aaaSDarrick J. Wong 22afc51aaaSDarrick J. Wong #include "../internal.h" 23afc51aaaSDarrick J. Wong 24ab08b01eSChristoph Hellwig /* 2595c4cd05SMatthew Wilcox (Oracle) * Structure allocated for each folio when block size < folio size 2695c4cd05SMatthew Wilcox (Oracle) * to track sub-folio uptodate status and I/O completions. 27ab08b01eSChristoph Hellwig */ 28ab08b01eSChristoph Hellwig struct iomap_page { 297d636676SMatthew Wilcox (Oracle) atomic_t read_bytes_pending; 300fb2d720SMatthew Wilcox (Oracle) atomic_t write_bytes_pending; 311cea335dSChristoph Hellwig spinlock_t uptodate_lock; 320a195b91SMatthew Wilcox (Oracle) unsigned long uptodate[]; 33ab08b01eSChristoph Hellwig }; 34ab08b01eSChristoph Hellwig 3595c4cd05SMatthew Wilcox (Oracle) static inline struct iomap_page *to_iomap_page(struct folio *folio) 36ab08b01eSChristoph Hellwig { 3795c4cd05SMatthew Wilcox (Oracle) if (folio_test_private(folio)) 3895c4cd05SMatthew Wilcox (Oracle) return folio_get_private(folio); 39ab08b01eSChristoph Hellwig return NULL; 40ab08b01eSChristoph Hellwig } 41ab08b01eSChristoph Hellwig 42598ecfbaSChristoph Hellwig static struct bio_set iomap_ioend_bioset; 43598ecfbaSChristoph Hellwig 44afc51aaaSDarrick J. Wong static struct iomap_page * 45435d44b3SMatthew Wilcox (Oracle) iomap_page_create(struct inode *inode, struct folio *folio) 46afc51aaaSDarrick J. Wong { 4795c4cd05SMatthew Wilcox (Oracle) struct iomap_page *iop = to_iomap_page(folio); 48435d44b3SMatthew Wilcox (Oracle) unsigned int nr_blocks = i_blocks_per_folio(inode, folio); 49afc51aaaSDarrick J. Wong 500a195b91SMatthew Wilcox (Oracle) if (iop || nr_blocks <= 1) 51afc51aaaSDarrick J. Wong return iop; 52afc51aaaSDarrick J. Wong 530a195b91SMatthew Wilcox (Oracle) iop = kzalloc(struct_size(iop, uptodate, BITS_TO_LONGS(nr_blocks)), 540a195b91SMatthew Wilcox (Oracle) GFP_NOFS | __GFP_NOFAIL); 551cea335dSChristoph Hellwig spin_lock_init(&iop->uptodate_lock); 56435d44b3SMatthew Wilcox (Oracle) if (folio_test_uptodate(folio)) 574595a298SMatthew Wilcox (Oracle) bitmap_fill(iop->uptodate, nr_blocks); 58435d44b3SMatthew Wilcox (Oracle) folio_attach_private(folio, iop); 59afc51aaaSDarrick J. Wong return iop; 60afc51aaaSDarrick J. Wong } 61afc51aaaSDarrick J. Wong 62c46e8324SMatthew Wilcox (Oracle) static void iomap_page_release(struct folio *folio) 63afc51aaaSDarrick J. Wong { 64c46e8324SMatthew Wilcox (Oracle) struct iomap_page *iop = folio_detach_private(folio); 65c46e8324SMatthew Wilcox (Oracle) struct inode *inode = folio->mapping->host; 66c46e8324SMatthew Wilcox (Oracle) unsigned int nr_blocks = i_blocks_per_folio(inode, folio); 67afc51aaaSDarrick J. Wong 68afc51aaaSDarrick J. Wong if (!iop) 69afc51aaaSDarrick J. Wong return; 707d636676SMatthew Wilcox (Oracle) WARN_ON_ONCE(atomic_read(&iop->read_bytes_pending)); 710fb2d720SMatthew Wilcox (Oracle) WARN_ON_ONCE(atomic_read(&iop->write_bytes_pending)); 720a195b91SMatthew Wilcox (Oracle) WARN_ON_ONCE(bitmap_full(iop->uptodate, nr_blocks) != 73c46e8324SMatthew Wilcox (Oracle) folio_test_uptodate(folio)); 74afc51aaaSDarrick J. Wong kfree(iop); 75afc51aaaSDarrick J. Wong } 76afc51aaaSDarrick J. Wong 77afc51aaaSDarrick J. Wong /* 78*431c0566SMatthew Wilcox (Oracle) * Calculate the range inside the folio that we actually need to read. 79afc51aaaSDarrick J. Wong */ 80*431c0566SMatthew Wilcox (Oracle) static void iomap_adjust_read_range(struct inode *inode, struct folio *folio, 81*431c0566SMatthew Wilcox (Oracle) loff_t *pos, loff_t length, size_t *offp, size_t *lenp) 82afc51aaaSDarrick J. Wong { 83*431c0566SMatthew Wilcox (Oracle) struct iomap_page *iop = to_iomap_page(folio); 84afc51aaaSDarrick J. Wong loff_t orig_pos = *pos; 85afc51aaaSDarrick J. Wong loff_t isize = i_size_read(inode); 86afc51aaaSDarrick J. Wong unsigned block_bits = inode->i_blkbits; 87afc51aaaSDarrick J. Wong unsigned block_size = (1 << block_bits); 88*431c0566SMatthew Wilcox (Oracle) size_t poff = offset_in_folio(folio, *pos); 89*431c0566SMatthew Wilcox (Oracle) size_t plen = min_t(loff_t, folio_size(folio) - poff, length); 90afc51aaaSDarrick J. Wong unsigned first = poff >> block_bits; 91afc51aaaSDarrick J. Wong unsigned last = (poff + plen - 1) >> block_bits; 92afc51aaaSDarrick J. Wong 93afc51aaaSDarrick J. Wong /* 94f1f264b4SAndreas Gruenbacher * If the block size is smaller than the page size, we need to check the 95afc51aaaSDarrick J. Wong * per-block uptodate status and adjust the offset and length if needed 96afc51aaaSDarrick J. Wong * to avoid reading in already uptodate ranges. 97afc51aaaSDarrick J. Wong */ 98afc51aaaSDarrick J. Wong if (iop) { 99afc51aaaSDarrick J. Wong unsigned int i; 100afc51aaaSDarrick J. Wong 101afc51aaaSDarrick J. Wong /* move forward for each leading block marked uptodate */ 102afc51aaaSDarrick J. Wong for (i = first; i <= last; i++) { 103afc51aaaSDarrick J. Wong if (!test_bit(i, iop->uptodate)) 104afc51aaaSDarrick J. Wong break; 105afc51aaaSDarrick J. Wong *pos += block_size; 106afc51aaaSDarrick J. Wong poff += block_size; 107afc51aaaSDarrick J. Wong plen -= block_size; 108afc51aaaSDarrick J. Wong first++; 109afc51aaaSDarrick J. Wong } 110afc51aaaSDarrick J. Wong 111afc51aaaSDarrick J. Wong /* truncate len if we find any trailing uptodate block(s) */ 112afc51aaaSDarrick J. Wong for ( ; i <= last; i++) { 113afc51aaaSDarrick J. Wong if (test_bit(i, iop->uptodate)) { 114afc51aaaSDarrick J. Wong plen -= (last - i + 1) * block_size; 115afc51aaaSDarrick J. Wong last = i - 1; 116afc51aaaSDarrick J. Wong break; 117afc51aaaSDarrick J. Wong } 118afc51aaaSDarrick J. Wong } 119afc51aaaSDarrick J. Wong } 120afc51aaaSDarrick J. Wong 121afc51aaaSDarrick J. Wong /* 122f1f264b4SAndreas Gruenbacher * If the extent spans the block that contains the i_size, we need to 123afc51aaaSDarrick J. Wong * handle both halves separately so that we properly zero data in the 124afc51aaaSDarrick J. Wong * page cache for blocks that are entirely outside of i_size. 125afc51aaaSDarrick J. Wong */ 126afc51aaaSDarrick J. Wong if (orig_pos <= isize && orig_pos + length > isize) { 127*431c0566SMatthew Wilcox (Oracle) unsigned end = offset_in_folio(folio, isize - 1) >> block_bits; 128afc51aaaSDarrick J. Wong 129afc51aaaSDarrick J. Wong if (first <= end && last > end) 130afc51aaaSDarrick J. Wong plen -= (last - end) * block_size; 131afc51aaaSDarrick J. Wong } 132afc51aaaSDarrick J. Wong 133afc51aaaSDarrick J. Wong *offp = poff; 134afc51aaaSDarrick J. Wong *lenp = plen; 135afc51aaaSDarrick J. Wong } 136afc51aaaSDarrick J. Wong 137*431c0566SMatthew Wilcox (Oracle) static void iomap_iop_set_range_uptodate(struct folio *folio, 138*431c0566SMatthew Wilcox (Oracle) struct iomap_page *iop, size_t off, size_t len) 139afc51aaaSDarrick J. Wong { 140*431c0566SMatthew Wilcox (Oracle) struct inode *inode = folio->mapping->host; 141afc51aaaSDarrick J. Wong unsigned first = off >> inode->i_blkbits; 142afc51aaaSDarrick J. Wong unsigned last = (off + len - 1) >> inode->i_blkbits; 1431cea335dSChristoph Hellwig unsigned long flags; 144afc51aaaSDarrick J. Wong 1451cea335dSChristoph Hellwig spin_lock_irqsave(&iop->uptodate_lock, flags); 146b21866f5SMatthew Wilcox (Oracle) bitmap_set(iop->uptodate, first, last - first + 1); 147*431c0566SMatthew Wilcox (Oracle) if (bitmap_full(iop->uptodate, i_blocks_per_folio(inode, folio))) 148*431c0566SMatthew Wilcox (Oracle) folio_mark_uptodate(folio); 1491cea335dSChristoph Hellwig spin_unlock_irqrestore(&iop->uptodate_lock, flags); 150afc51aaaSDarrick J. Wong } 151afc51aaaSDarrick J. Wong 152*431c0566SMatthew Wilcox (Oracle) static void iomap_set_range_uptodate(struct folio *folio, 153*431c0566SMatthew Wilcox (Oracle) struct iomap_page *iop, size_t off, size_t len) 1541cea335dSChristoph Hellwig { 155*431c0566SMatthew Wilcox (Oracle) if (folio_test_error(folio)) 1561cea335dSChristoph Hellwig return; 1571cea335dSChristoph Hellwig 158cd1e5afeSMatthew Wilcox (Oracle) if (iop) 159*431c0566SMatthew Wilcox (Oracle) iomap_iop_set_range_uptodate(folio, iop, off, len); 1601cea335dSChristoph Hellwig else 161*431c0566SMatthew Wilcox (Oracle) folio_mark_uptodate(folio); 162afc51aaaSDarrick J. Wong } 163afc51aaaSDarrick J. Wong 1648ffd74e9SMatthew Wilcox (Oracle) static void iomap_finish_folio_read(struct folio *folio, size_t offset, 1658ffd74e9SMatthew Wilcox (Oracle) size_t len, int error) 166afc51aaaSDarrick J. Wong { 16795c4cd05SMatthew Wilcox (Oracle) struct iomap_page *iop = to_iomap_page(folio); 168afc51aaaSDarrick J. Wong 169afc51aaaSDarrick J. Wong if (unlikely(error)) { 1708ffd74e9SMatthew Wilcox (Oracle) folio_clear_uptodate(folio); 1718ffd74e9SMatthew Wilcox (Oracle) folio_set_error(folio); 172afc51aaaSDarrick J. Wong } else { 173*431c0566SMatthew Wilcox (Oracle) iomap_set_range_uptodate(folio, iop, offset, len); 174afc51aaaSDarrick J. Wong } 175afc51aaaSDarrick J. Wong 1768ffd74e9SMatthew Wilcox (Oracle) if (!iop || atomic_sub_and_test(len, &iop->read_bytes_pending)) 1778ffd74e9SMatthew Wilcox (Oracle) folio_unlock(folio); 178afc51aaaSDarrick J. Wong } 179afc51aaaSDarrick J. Wong 1808ffd74e9SMatthew Wilcox (Oracle) static void iomap_read_end_io(struct bio *bio) 181afc51aaaSDarrick J. Wong { 182afc51aaaSDarrick J. Wong int error = blk_status_to_errno(bio->bi_status); 1838ffd74e9SMatthew Wilcox (Oracle) struct folio_iter fi; 184afc51aaaSDarrick J. Wong 1858ffd74e9SMatthew Wilcox (Oracle) bio_for_each_folio_all(fi, bio) 1868ffd74e9SMatthew Wilcox (Oracle) iomap_finish_folio_read(fi.folio, fi.offset, fi.length, error); 187afc51aaaSDarrick J. Wong bio_put(bio); 188afc51aaaSDarrick J. Wong } 189afc51aaaSDarrick J. Wong 190afc51aaaSDarrick J. Wong struct iomap_readpage_ctx { 191afc51aaaSDarrick J. Wong struct page *cur_page; 192afc51aaaSDarrick J. Wong bool cur_page_in_bio; 193afc51aaaSDarrick J. Wong struct bio *bio; 1949d24a13aSMatthew Wilcox (Oracle) struct readahead_control *rac; 195afc51aaaSDarrick J. Wong }; 196afc51aaaSDarrick J. Wong 1975ad448ceSAndreas Gruenbacher /** 1985ad448ceSAndreas Gruenbacher * iomap_read_inline_data - copy inline data into the page cache 1995ad448ceSAndreas Gruenbacher * @iter: iteration structure 2005ad448ceSAndreas Gruenbacher * @page: page to copy to 2015ad448ceSAndreas Gruenbacher * 2025ad448ceSAndreas Gruenbacher * Copy the inline data in @iter into @page and zero out the rest of the page. 2035ad448ceSAndreas Gruenbacher * Only a single IOMAP_INLINE extent is allowed at the end of each file. 2045ad448ceSAndreas Gruenbacher * Returns zero for success to complete the read, or the usual negative errno. 2055ad448ceSAndreas Gruenbacher */ 2065ad448ceSAndreas Gruenbacher static int iomap_read_inline_data(const struct iomap_iter *iter, 2071b5c1e36SChristoph Hellwig struct page *page) 208afc51aaaSDarrick J. Wong { 209435d44b3SMatthew Wilcox (Oracle) struct folio *folio = page_folio(page); 210cd1e5afeSMatthew Wilcox (Oracle) struct iomap_page *iop; 211fad0a1abSChristoph Hellwig const struct iomap *iomap = iomap_iter_srcmap(iter); 2121b5c1e36SChristoph Hellwig size_t size = i_size_read(iter->inode) - iomap->offset; 213b405435bSMatthew Wilcox (Oracle) size_t poff = offset_in_page(iomap->offset); 214*431c0566SMatthew Wilcox (Oracle) size_t offset = offset_in_folio(folio, iomap->offset); 215afc51aaaSDarrick J. Wong void *addr; 216afc51aaaSDarrick J. Wong 217afc51aaaSDarrick J. Wong if (PageUptodate(page)) 2185ad448ceSAndreas Gruenbacher return 0; 219afc51aaaSDarrick J. Wong 220ae44f9c2SMatthew Wilcox (Oracle) if (WARN_ON_ONCE(size > PAGE_SIZE - poff)) 221ae44f9c2SMatthew Wilcox (Oracle) return -EIO; 22269f4a26cSGao Xiang if (WARN_ON_ONCE(size > PAGE_SIZE - 22369f4a26cSGao Xiang offset_in_page(iomap->inline_data))) 22469f4a26cSGao Xiang return -EIO; 22569f4a26cSGao Xiang if (WARN_ON_ONCE(size > iomap->length)) 22669f4a26cSGao Xiang return -EIO; 227*431c0566SMatthew Wilcox (Oracle) if (offset > 0) 228cd1e5afeSMatthew Wilcox (Oracle) iop = iomap_page_create(iter->inode, folio); 229cd1e5afeSMatthew Wilcox (Oracle) else 230cd1e5afeSMatthew Wilcox (Oracle) iop = to_iomap_page(folio); 231afc51aaaSDarrick J. Wong 232ab069d5fSMatthew Wilcox (Oracle) addr = kmap_local_page(page) + poff; 233afc51aaaSDarrick J. Wong memcpy(addr, iomap->inline_data, size); 234b405435bSMatthew Wilcox (Oracle) memset(addr + size, 0, PAGE_SIZE - poff - size); 235ab069d5fSMatthew Wilcox (Oracle) kunmap_local(addr); 236*431c0566SMatthew Wilcox (Oracle) iomap_set_range_uptodate(folio, iop, offset, PAGE_SIZE - poff); 2375ad448ceSAndreas Gruenbacher return 0; 238afc51aaaSDarrick J. Wong } 239afc51aaaSDarrick J. Wong 240fad0a1abSChristoph Hellwig static inline bool iomap_block_needs_zeroing(const struct iomap_iter *iter, 2411b5c1e36SChristoph Hellwig loff_t pos) 242009d8d84SChristoph Hellwig { 243fad0a1abSChristoph Hellwig const struct iomap *srcmap = iomap_iter_srcmap(iter); 2441b5c1e36SChristoph Hellwig 2451b5c1e36SChristoph Hellwig return srcmap->type != IOMAP_MAPPED || 2461b5c1e36SChristoph Hellwig (srcmap->flags & IOMAP_F_NEW) || 2471b5c1e36SChristoph Hellwig pos >= i_size_read(iter->inode); 248009d8d84SChristoph Hellwig } 249009d8d84SChristoph Hellwig 250fad0a1abSChristoph Hellwig static loff_t iomap_readpage_iter(const struct iomap_iter *iter, 251f6d48000SChristoph Hellwig struct iomap_readpage_ctx *ctx, loff_t offset) 252afc51aaaSDarrick J. Wong { 253fad0a1abSChristoph Hellwig const struct iomap *iomap = &iter->iomap; 254f6d48000SChristoph Hellwig loff_t pos = iter->pos + offset; 255f6d48000SChristoph Hellwig loff_t length = iomap_length(iter) - offset; 256afc51aaaSDarrick J. Wong struct page *page = ctx->cur_page; 257435d44b3SMatthew Wilcox (Oracle) struct folio *folio = page_folio(page); 258637d3375SAndreas Gruenbacher struct iomap_page *iop; 259afc51aaaSDarrick J. Wong loff_t orig_pos = pos; 260*431c0566SMatthew Wilcox (Oracle) size_t poff, plen; 261afc51aaaSDarrick J. Wong sector_t sector; 262afc51aaaSDarrick J. Wong 2635ad448ceSAndreas Gruenbacher if (iomap->type == IOMAP_INLINE) 2645ad448ceSAndreas Gruenbacher return iomap_read_inline_data(iter, page); 265afc51aaaSDarrick J. Wong 266afc51aaaSDarrick J. Wong /* zero post-eof blocks as the page may be mapped */ 267435d44b3SMatthew Wilcox (Oracle) iop = iomap_page_create(iter->inode, folio); 268*431c0566SMatthew Wilcox (Oracle) iomap_adjust_read_range(iter->inode, folio, &pos, length, &poff, &plen); 269afc51aaaSDarrick J. Wong if (plen == 0) 270afc51aaaSDarrick J. Wong goto done; 271afc51aaaSDarrick J. Wong 2721b5c1e36SChristoph Hellwig if (iomap_block_needs_zeroing(iter, pos)) { 273*431c0566SMatthew Wilcox (Oracle) folio_zero_range(folio, poff, plen); 274*431c0566SMatthew Wilcox (Oracle) iomap_set_range_uptodate(folio, iop, poff, plen); 275afc51aaaSDarrick J. Wong goto done; 276afc51aaaSDarrick J. Wong } 277afc51aaaSDarrick J. Wong 278afc51aaaSDarrick J. Wong ctx->cur_page_in_bio = true; 2797d636676SMatthew Wilcox (Oracle) if (iop) 2807d636676SMatthew Wilcox (Oracle) atomic_add(plen, &iop->read_bytes_pending); 281afc51aaaSDarrick J. Wong 282afc51aaaSDarrick J. Wong sector = iomap_sector(iomap, pos); 283d0364f94SChristoph Hellwig if (!ctx->bio || 284d0364f94SChristoph Hellwig bio_end_sector(ctx->bio) != sector || 285*431c0566SMatthew Wilcox (Oracle) !bio_add_folio(ctx->bio, folio, plen, poff)) { 286afc51aaaSDarrick J. Wong gfp_t gfp = mapping_gfp_constraint(page->mapping, GFP_KERNEL); 287457df33eSMatthew Wilcox (Oracle) gfp_t orig_gfp = gfp; 2885f7136dbSMatthew Wilcox (Oracle) unsigned int nr_vecs = DIV_ROUND_UP(length, PAGE_SIZE); 289afc51aaaSDarrick J. Wong 290afc51aaaSDarrick J. Wong if (ctx->bio) 291afc51aaaSDarrick J. Wong submit_bio(ctx->bio); 292afc51aaaSDarrick J. Wong 2939d24a13aSMatthew Wilcox (Oracle) if (ctx->rac) /* same as readahead_gfp_mask */ 294afc51aaaSDarrick J. Wong gfp |= __GFP_NORETRY | __GFP_NOWARN; 2955f7136dbSMatthew Wilcox (Oracle) ctx->bio = bio_alloc(gfp, bio_max_segs(nr_vecs)); 296457df33eSMatthew Wilcox (Oracle) /* 297457df33eSMatthew Wilcox (Oracle) * If the bio_alloc fails, try it again for a single page to 298457df33eSMatthew Wilcox (Oracle) * avoid having to deal with partial page reads. This emulates 299457df33eSMatthew Wilcox (Oracle) * what do_mpage_readpage does. 300457df33eSMatthew Wilcox (Oracle) */ 301457df33eSMatthew Wilcox (Oracle) if (!ctx->bio) 302457df33eSMatthew Wilcox (Oracle) ctx->bio = bio_alloc(orig_gfp, 1); 303afc51aaaSDarrick J. Wong ctx->bio->bi_opf = REQ_OP_READ; 3049d24a13aSMatthew Wilcox (Oracle) if (ctx->rac) 305afc51aaaSDarrick J. Wong ctx->bio->bi_opf |= REQ_RAHEAD; 306afc51aaaSDarrick J. Wong ctx->bio->bi_iter.bi_sector = sector; 307afc51aaaSDarrick J. Wong bio_set_dev(ctx->bio, iomap->bdev); 308afc51aaaSDarrick J. Wong ctx->bio->bi_end_io = iomap_read_end_io; 309*431c0566SMatthew Wilcox (Oracle) bio_add_folio(ctx->bio, folio, plen, poff); 310afc51aaaSDarrick J. Wong } 311*431c0566SMatthew Wilcox (Oracle) 312afc51aaaSDarrick J. Wong done: 313afc51aaaSDarrick J. Wong /* 314afc51aaaSDarrick J. Wong * Move the caller beyond our range so that it keeps making progress. 315f1f264b4SAndreas Gruenbacher * For that, we have to include any leading non-uptodate ranges, but 316afc51aaaSDarrick J. Wong * we can skip trailing ones as they will be handled in the next 317afc51aaaSDarrick J. Wong * iteration. 318afc51aaaSDarrick J. Wong */ 319afc51aaaSDarrick J. Wong return pos - orig_pos + plen; 320afc51aaaSDarrick J. Wong } 321afc51aaaSDarrick J. Wong 322afc51aaaSDarrick J. Wong int 323afc51aaaSDarrick J. Wong iomap_readpage(struct page *page, const struct iomap_ops *ops) 324afc51aaaSDarrick J. Wong { 325f6d48000SChristoph Hellwig struct iomap_iter iter = { 326f6d48000SChristoph Hellwig .inode = page->mapping->host, 327f6d48000SChristoph Hellwig .pos = page_offset(page), 328f6d48000SChristoph Hellwig .len = PAGE_SIZE, 329f6d48000SChristoph Hellwig }; 330f6d48000SChristoph Hellwig struct iomap_readpage_ctx ctx = { 331f6d48000SChristoph Hellwig .cur_page = page, 332f6d48000SChristoph Hellwig }; 333f6d48000SChristoph Hellwig int ret; 334afc51aaaSDarrick J. Wong 3359e91c572SChristoph Hellwig trace_iomap_readpage(page->mapping->host, 1); 3369e91c572SChristoph Hellwig 337f6d48000SChristoph Hellwig while ((ret = iomap_iter(&iter, ops)) > 0) 338f6d48000SChristoph Hellwig iter.processed = iomap_readpage_iter(&iter, &ctx, 0); 339f6d48000SChristoph Hellwig 340f6d48000SChristoph Hellwig if (ret < 0) 341afc51aaaSDarrick J. Wong SetPageError(page); 342afc51aaaSDarrick J. Wong 343afc51aaaSDarrick J. Wong if (ctx.bio) { 344afc51aaaSDarrick J. Wong submit_bio(ctx.bio); 345afc51aaaSDarrick J. Wong WARN_ON_ONCE(!ctx.cur_page_in_bio); 346afc51aaaSDarrick J. Wong } else { 347afc51aaaSDarrick J. Wong WARN_ON_ONCE(ctx.cur_page_in_bio); 348afc51aaaSDarrick J. Wong unlock_page(page); 349afc51aaaSDarrick J. Wong } 350afc51aaaSDarrick J. Wong 351afc51aaaSDarrick J. Wong /* 352f1f264b4SAndreas Gruenbacher * Just like mpage_readahead and block_read_full_page, we always 353afc51aaaSDarrick J. Wong * return 0 and just mark the page as PageError on errors. This 354f1f264b4SAndreas Gruenbacher * should be cleaned up throughout the stack eventually. 355afc51aaaSDarrick J. Wong */ 356afc51aaaSDarrick J. Wong return 0; 357afc51aaaSDarrick J. Wong } 358afc51aaaSDarrick J. Wong EXPORT_SYMBOL_GPL(iomap_readpage); 359afc51aaaSDarrick J. Wong 360fad0a1abSChristoph Hellwig static loff_t iomap_readahead_iter(const struct iomap_iter *iter, 361f6d48000SChristoph Hellwig struct iomap_readpage_ctx *ctx) 362afc51aaaSDarrick J. Wong { 363f6d48000SChristoph Hellwig loff_t length = iomap_length(iter); 364afc51aaaSDarrick J. Wong loff_t done, ret; 365afc51aaaSDarrick J. Wong 366afc51aaaSDarrick J. Wong for (done = 0; done < length; done += ret) { 367f6d48000SChristoph Hellwig if (ctx->cur_page && offset_in_page(iter->pos + done) == 0) { 368afc51aaaSDarrick J. Wong if (!ctx->cur_page_in_bio) 369afc51aaaSDarrick J. Wong unlock_page(ctx->cur_page); 370afc51aaaSDarrick J. Wong put_page(ctx->cur_page); 371afc51aaaSDarrick J. Wong ctx->cur_page = NULL; 372afc51aaaSDarrick J. Wong } 373afc51aaaSDarrick J. Wong if (!ctx->cur_page) { 3749d24a13aSMatthew Wilcox (Oracle) ctx->cur_page = readahead_page(ctx->rac); 375afc51aaaSDarrick J. Wong ctx->cur_page_in_bio = false; 376afc51aaaSDarrick J. Wong } 377f6d48000SChristoph Hellwig ret = iomap_readpage_iter(iter, ctx, done); 378d8af404fSAndreas Gruenbacher if (ret <= 0) 379d8af404fSAndreas Gruenbacher return ret; 380afc51aaaSDarrick J. Wong } 381afc51aaaSDarrick J. Wong 382afc51aaaSDarrick J. Wong return done; 383afc51aaaSDarrick J. Wong } 384afc51aaaSDarrick J. Wong 3859d24a13aSMatthew Wilcox (Oracle) /** 3869d24a13aSMatthew Wilcox (Oracle) * iomap_readahead - Attempt to read pages from a file. 3879d24a13aSMatthew Wilcox (Oracle) * @rac: Describes the pages to be read. 3889d24a13aSMatthew Wilcox (Oracle) * @ops: The operations vector for the filesystem. 3899d24a13aSMatthew Wilcox (Oracle) * 3909d24a13aSMatthew Wilcox (Oracle) * This function is for filesystems to call to implement their readahead 3919d24a13aSMatthew Wilcox (Oracle) * address_space operation. 3929d24a13aSMatthew Wilcox (Oracle) * 3939d24a13aSMatthew Wilcox (Oracle) * Context: The @ops callbacks may submit I/O (eg to read the addresses of 3949d24a13aSMatthew Wilcox (Oracle) * blocks from disc), and may wait for it. The caller may be trying to 3959d24a13aSMatthew Wilcox (Oracle) * access a different page, and so sleeping excessively should be avoided. 3969d24a13aSMatthew Wilcox (Oracle) * It may allocate memory, but should avoid costly allocations. This 3979d24a13aSMatthew Wilcox (Oracle) * function is called with memalloc_nofs set, so allocations will not cause 3989d24a13aSMatthew Wilcox (Oracle) * the filesystem to be reentered. 3999d24a13aSMatthew Wilcox (Oracle) */ 4009d24a13aSMatthew Wilcox (Oracle) void iomap_readahead(struct readahead_control *rac, const struct iomap_ops *ops) 401afc51aaaSDarrick J. Wong { 402f6d48000SChristoph Hellwig struct iomap_iter iter = { 403f6d48000SChristoph Hellwig .inode = rac->mapping->host, 404f6d48000SChristoph Hellwig .pos = readahead_pos(rac), 405f6d48000SChristoph Hellwig .len = readahead_length(rac), 406f6d48000SChristoph Hellwig }; 407afc51aaaSDarrick J. Wong struct iomap_readpage_ctx ctx = { 4089d24a13aSMatthew Wilcox (Oracle) .rac = rac, 409afc51aaaSDarrick J. Wong }; 410afc51aaaSDarrick J. Wong 411f6d48000SChristoph Hellwig trace_iomap_readahead(rac->mapping->host, readahead_count(rac)); 4129e91c572SChristoph Hellwig 413f6d48000SChristoph Hellwig while (iomap_iter(&iter, ops) > 0) 414f6d48000SChristoph Hellwig iter.processed = iomap_readahead_iter(&iter, &ctx); 4159d24a13aSMatthew Wilcox (Oracle) 416afc51aaaSDarrick J. Wong if (ctx.bio) 417afc51aaaSDarrick J. Wong submit_bio(ctx.bio); 418afc51aaaSDarrick J. Wong if (ctx.cur_page) { 419afc51aaaSDarrick J. Wong if (!ctx.cur_page_in_bio) 420afc51aaaSDarrick J. Wong unlock_page(ctx.cur_page); 421afc51aaaSDarrick J. Wong put_page(ctx.cur_page); 422afc51aaaSDarrick J. Wong } 423afc51aaaSDarrick J. Wong } 4249d24a13aSMatthew Wilcox (Oracle) EXPORT_SYMBOL_GPL(iomap_readahead); 425afc51aaaSDarrick J. Wong 426afc51aaaSDarrick J. Wong /* 427afc51aaaSDarrick J. Wong * iomap_is_partially_uptodate checks whether blocks within a page are 428afc51aaaSDarrick J. Wong * uptodate or not. 429afc51aaaSDarrick J. Wong * 430afc51aaaSDarrick J. Wong * Returns true if all blocks which correspond to a file portion 431afc51aaaSDarrick J. Wong * we want to read within the page are uptodate. 432afc51aaaSDarrick J. Wong */ 433afc51aaaSDarrick J. Wong int 434afc51aaaSDarrick J. Wong iomap_is_partially_uptodate(struct page *page, unsigned long from, 435afc51aaaSDarrick J. Wong unsigned long count) 436afc51aaaSDarrick J. Wong { 43795c4cd05SMatthew Wilcox (Oracle) struct folio *folio = page_folio(page); 43895c4cd05SMatthew Wilcox (Oracle) struct iomap_page *iop = to_iomap_page(folio); 439afc51aaaSDarrick J. Wong struct inode *inode = page->mapping->host; 440afc51aaaSDarrick J. Wong unsigned len, first, last; 441afc51aaaSDarrick J. Wong unsigned i; 442afc51aaaSDarrick J. Wong 443afc51aaaSDarrick J. Wong /* Limit range to one page */ 444afc51aaaSDarrick J. Wong len = min_t(unsigned, PAGE_SIZE - from, count); 445afc51aaaSDarrick J. Wong 446afc51aaaSDarrick J. Wong /* First and last blocks in range within page */ 447afc51aaaSDarrick J. Wong first = from >> inode->i_blkbits; 448afc51aaaSDarrick J. Wong last = (from + len - 1) >> inode->i_blkbits; 449afc51aaaSDarrick J. Wong 450afc51aaaSDarrick J. Wong if (iop) { 451afc51aaaSDarrick J. Wong for (i = first; i <= last; i++) 452afc51aaaSDarrick J. Wong if (!test_bit(i, iop->uptodate)) 453afc51aaaSDarrick J. Wong return 0; 454afc51aaaSDarrick J. Wong return 1; 455afc51aaaSDarrick J. Wong } 456afc51aaaSDarrick J. Wong 457afc51aaaSDarrick J. Wong return 0; 458afc51aaaSDarrick J. Wong } 459afc51aaaSDarrick J. Wong EXPORT_SYMBOL_GPL(iomap_is_partially_uptodate); 460afc51aaaSDarrick J. Wong 461afc51aaaSDarrick J. Wong int 462afc51aaaSDarrick J. Wong iomap_releasepage(struct page *page, gfp_t gfp_mask) 463afc51aaaSDarrick J. Wong { 464c46e8324SMatthew Wilcox (Oracle) struct folio *folio = page_folio(page); 465c46e8324SMatthew Wilcox (Oracle) 46639f16c83SMatthew Wilcox (Oracle) trace_iomap_releasepage(folio->mapping->host, folio_pos(folio), 46739f16c83SMatthew Wilcox (Oracle) folio_size(folio)); 4689e91c572SChristoph Hellwig 469afc51aaaSDarrick J. Wong /* 470afc51aaaSDarrick J. Wong * mm accommodates an old ext3 case where clean pages might not have had 471afc51aaaSDarrick J. Wong * the dirty bit cleared. Thus, it can send actual dirty pages to 472f1f264b4SAndreas Gruenbacher * ->releasepage() via shrink_active_list(); skip those here. 473afc51aaaSDarrick J. Wong */ 47439f16c83SMatthew Wilcox (Oracle) if (folio_test_dirty(folio) || folio_test_writeback(folio)) 475afc51aaaSDarrick J. Wong return 0; 476c46e8324SMatthew Wilcox (Oracle) iomap_page_release(folio); 477afc51aaaSDarrick J. Wong return 1; 478afc51aaaSDarrick J. Wong } 479afc51aaaSDarrick J. Wong EXPORT_SYMBOL_GPL(iomap_releasepage); 480afc51aaaSDarrick J. Wong 4818306a5f5SMatthew Wilcox (Oracle) void iomap_invalidate_folio(struct folio *folio, size_t offset, size_t len) 482afc51aaaSDarrick J. Wong { 4838306a5f5SMatthew Wilcox (Oracle) trace_iomap_invalidatepage(folio->mapping->host, offset, len); 4849e91c572SChristoph Hellwig 485afc51aaaSDarrick J. Wong /* 486f1f264b4SAndreas Gruenbacher * If we're invalidating the entire page, clear the dirty state from it 487afc51aaaSDarrick J. Wong * and release it to avoid unnecessary buildup of the LRU. 488afc51aaaSDarrick J. Wong */ 4898306a5f5SMatthew Wilcox (Oracle) if (offset == 0 && len == folio_size(folio)) { 4908306a5f5SMatthew Wilcox (Oracle) WARN_ON_ONCE(folio_test_writeback(folio)); 4918306a5f5SMatthew Wilcox (Oracle) folio_cancel_dirty(folio); 492c46e8324SMatthew Wilcox (Oracle) iomap_page_release(folio); 493afc51aaaSDarrick J. Wong } 494afc51aaaSDarrick J. Wong } 4958306a5f5SMatthew Wilcox (Oracle) EXPORT_SYMBOL_GPL(iomap_invalidate_folio); 4968306a5f5SMatthew Wilcox (Oracle) 4978306a5f5SMatthew Wilcox (Oracle) void iomap_invalidatepage(struct page *page, unsigned int offset, 4988306a5f5SMatthew Wilcox (Oracle) unsigned int len) 4998306a5f5SMatthew Wilcox (Oracle) { 5008306a5f5SMatthew Wilcox (Oracle) iomap_invalidate_folio(page_folio(page), offset, len); 5018306a5f5SMatthew Wilcox (Oracle) } 502afc51aaaSDarrick J. Wong EXPORT_SYMBOL_GPL(iomap_invalidatepage); 503afc51aaaSDarrick J. Wong 504afc51aaaSDarrick J. Wong #ifdef CONFIG_MIGRATION 505afc51aaaSDarrick J. Wong int 506afc51aaaSDarrick J. Wong iomap_migrate_page(struct address_space *mapping, struct page *newpage, 507afc51aaaSDarrick J. Wong struct page *page, enum migrate_mode mode) 508afc51aaaSDarrick J. Wong { 509afc51aaaSDarrick J. Wong int ret; 510afc51aaaSDarrick J. Wong 51126473f83SLinus Torvalds ret = migrate_page_move_mapping(mapping, newpage, page, 0); 512afc51aaaSDarrick J. Wong if (ret != MIGRATEPAGE_SUCCESS) 513afc51aaaSDarrick J. Wong return ret; 514afc51aaaSDarrick J. Wong 51558aeb731SGuoqing Jiang if (page_has_private(page)) 51658aeb731SGuoqing Jiang attach_page_private(newpage, detach_page_private(page)); 517afc51aaaSDarrick J. Wong 518afc51aaaSDarrick J. Wong if (mode != MIGRATE_SYNC_NO_COPY) 519afc51aaaSDarrick J. Wong migrate_page_copy(newpage, page); 520afc51aaaSDarrick J. Wong else 521afc51aaaSDarrick J. Wong migrate_page_states(newpage, page); 522afc51aaaSDarrick J. Wong return MIGRATEPAGE_SUCCESS; 523afc51aaaSDarrick J. Wong } 524afc51aaaSDarrick J. Wong EXPORT_SYMBOL_GPL(iomap_migrate_page); 525afc51aaaSDarrick J. Wong #endif /* CONFIG_MIGRATION */ 526afc51aaaSDarrick J. Wong 527afc51aaaSDarrick J. Wong static void 528afc51aaaSDarrick J. Wong iomap_write_failed(struct inode *inode, loff_t pos, unsigned len) 529afc51aaaSDarrick J. Wong { 530afc51aaaSDarrick J. Wong loff_t i_size = i_size_read(inode); 531afc51aaaSDarrick J. Wong 532afc51aaaSDarrick J. Wong /* 533afc51aaaSDarrick J. Wong * Only truncate newly allocated pages beyoned EOF, even if the 534afc51aaaSDarrick J. Wong * write started inside the existing inode size. 535afc51aaaSDarrick J. Wong */ 536afc51aaaSDarrick J. Wong if (pos + len > i_size) 537afc51aaaSDarrick J. Wong truncate_pagecache_range(inode, max(pos, i_size), pos + len); 538afc51aaaSDarrick J. Wong } 539afc51aaaSDarrick J. Wong 540*431c0566SMatthew Wilcox (Oracle) static int iomap_read_folio_sync(loff_t block_start, struct folio *folio, 541*431c0566SMatthew Wilcox (Oracle) size_t poff, size_t plen, const struct iomap *iomap) 542afc51aaaSDarrick J. Wong { 543afc51aaaSDarrick J. Wong struct bio_vec bvec; 544afc51aaaSDarrick J. Wong struct bio bio; 545afc51aaaSDarrick J. Wong 546afc51aaaSDarrick J. Wong bio_init(&bio, &bvec, 1); 547afc51aaaSDarrick J. Wong bio.bi_opf = REQ_OP_READ; 548afc51aaaSDarrick J. Wong bio.bi_iter.bi_sector = iomap_sector(iomap, block_start); 549afc51aaaSDarrick J. Wong bio_set_dev(&bio, iomap->bdev); 550*431c0566SMatthew Wilcox (Oracle) bio_add_folio(&bio, folio, plen, poff); 551afc51aaaSDarrick J. Wong return submit_bio_wait(&bio); 552afc51aaaSDarrick J. Wong } 553afc51aaaSDarrick J. Wong 554fad0a1abSChristoph Hellwig static int __iomap_write_begin(const struct iomap_iter *iter, loff_t pos, 555b74b1293SChristoph Hellwig unsigned len, struct page *page) 556afc51aaaSDarrick J. Wong { 557435d44b3SMatthew Wilcox (Oracle) struct folio *folio = page_folio(page); 558fad0a1abSChristoph Hellwig const struct iomap *srcmap = iomap_iter_srcmap(iter); 559435d44b3SMatthew Wilcox (Oracle) struct iomap_page *iop = iomap_page_create(iter->inode, folio); 5601b5c1e36SChristoph Hellwig loff_t block_size = i_blocksize(iter->inode); 5616cc19c5fSNikolay Borisov loff_t block_start = round_down(pos, block_size); 5626cc19c5fSNikolay Borisov loff_t block_end = round_up(pos + len, block_size); 563*431c0566SMatthew Wilcox (Oracle) size_t from = offset_in_folio(folio, pos), to = from + len; 564*431c0566SMatthew Wilcox (Oracle) size_t poff, plen; 565afc51aaaSDarrick J. Wong 566*431c0566SMatthew Wilcox (Oracle) if (folio_test_uptodate(folio)) 567afc51aaaSDarrick J. Wong return 0; 568*431c0566SMatthew Wilcox (Oracle) folio_clear_error(folio); 569afc51aaaSDarrick J. Wong 570afc51aaaSDarrick J. Wong do { 571*431c0566SMatthew Wilcox (Oracle) iomap_adjust_read_range(iter->inode, folio, &block_start, 572afc51aaaSDarrick J. Wong block_end - block_start, &poff, &plen); 573afc51aaaSDarrick J. Wong if (plen == 0) 574afc51aaaSDarrick J. Wong break; 575afc51aaaSDarrick J. Wong 576b74b1293SChristoph Hellwig if (!(iter->flags & IOMAP_UNSHARE) && 57732a38a49SChristoph Hellwig (from <= poff || from >= poff + plen) && 578d3b40439SChristoph Hellwig (to <= poff || to >= poff + plen)) 579d3b40439SChristoph Hellwig continue; 580d3b40439SChristoph Hellwig 5811b5c1e36SChristoph Hellwig if (iomap_block_needs_zeroing(iter, block_start)) { 582b74b1293SChristoph Hellwig if (WARN_ON_ONCE(iter->flags & IOMAP_UNSHARE)) 58332a38a49SChristoph Hellwig return -EIO; 584*431c0566SMatthew Wilcox (Oracle) folio_zero_segments(folio, poff, from, to, poff + plen); 58514284fedSMatthew Wilcox (Oracle) } else { 586*431c0566SMatthew Wilcox (Oracle) int status = iomap_read_folio_sync(block_start, folio, 58714284fedSMatthew Wilcox (Oracle) poff, plen, srcmap); 588d3b40439SChristoph Hellwig if (status) 589d3b40439SChristoph Hellwig return status; 59014284fedSMatthew Wilcox (Oracle) } 591*431c0566SMatthew Wilcox (Oracle) iomap_set_range_uptodate(folio, iop, poff, plen); 592afc51aaaSDarrick J. Wong } while ((block_start += plen) < block_end); 593afc51aaaSDarrick J. Wong 594d3b40439SChristoph Hellwig return 0; 595afc51aaaSDarrick J. Wong } 596afc51aaaSDarrick J. Wong 597fad0a1abSChristoph Hellwig static int iomap_write_begin_inline(const struct iomap_iter *iter, 5981b5c1e36SChristoph Hellwig struct page *page) 59969f4a26cSGao Xiang { 60069f4a26cSGao Xiang /* needs more work for the tailpacking case; disable for now */ 6011b5c1e36SChristoph Hellwig if (WARN_ON_ONCE(iomap_iter_srcmap(iter)->offset != 0)) 60269f4a26cSGao Xiang return -EIO; 6035ad448ceSAndreas Gruenbacher return iomap_read_inline_data(iter, page); 60469f4a26cSGao Xiang } 60569f4a26cSGao Xiang 606fad0a1abSChristoph Hellwig static int iomap_write_begin(const struct iomap_iter *iter, loff_t pos, 607fad0a1abSChristoph Hellwig unsigned len, struct page **pagep) 608afc51aaaSDarrick J. Wong { 6091b5c1e36SChristoph Hellwig const struct iomap_page_ops *page_ops = iter->iomap.page_ops; 610fad0a1abSChristoph Hellwig const struct iomap *srcmap = iomap_iter_srcmap(iter); 611afc51aaaSDarrick J. Wong struct page *page; 612d1bd0b4eSMatthew Wilcox (Oracle) struct folio *folio; 613afc51aaaSDarrick J. Wong int status = 0; 614afc51aaaSDarrick J. Wong 6151b5c1e36SChristoph Hellwig BUG_ON(pos + len > iter->iomap.offset + iter->iomap.length); 6161b5c1e36SChristoph Hellwig if (srcmap != &iter->iomap) 617c039b997SGoldwyn Rodrigues BUG_ON(pos + len > srcmap->offset + srcmap->length); 618afc51aaaSDarrick J. Wong 619afc51aaaSDarrick J. Wong if (fatal_signal_pending(current)) 620afc51aaaSDarrick J. Wong return -EINTR; 621afc51aaaSDarrick J. Wong 622afc51aaaSDarrick J. Wong if (page_ops && page_ops->page_prepare) { 6231b5c1e36SChristoph Hellwig status = page_ops->page_prepare(iter->inode, pos, len); 624afc51aaaSDarrick J. Wong if (status) 625afc51aaaSDarrick J. Wong return status; 626afc51aaaSDarrick J. Wong } 627afc51aaaSDarrick J. Wong 6281b5c1e36SChristoph Hellwig page = grab_cache_page_write_begin(iter->inode->i_mapping, 6291b5c1e36SChristoph Hellwig pos >> PAGE_SHIFT, AOP_FLAG_NOFS); 630afc51aaaSDarrick J. Wong if (!page) { 631afc51aaaSDarrick J. Wong status = -ENOMEM; 632afc51aaaSDarrick J. Wong goto out_no_page; 633afc51aaaSDarrick J. Wong } 634d1bd0b4eSMatthew Wilcox (Oracle) folio = page_folio(page); 635afc51aaaSDarrick J. Wong 636c039b997SGoldwyn Rodrigues if (srcmap->type == IOMAP_INLINE) 6371b5c1e36SChristoph Hellwig status = iomap_write_begin_inline(iter, page); 6381b5c1e36SChristoph Hellwig else if (srcmap->flags & IOMAP_F_BUFFER_HEAD) 639d1bd0b4eSMatthew Wilcox (Oracle) status = __block_write_begin_int(folio, pos, len, NULL, srcmap); 640afc51aaaSDarrick J. Wong else 641b74b1293SChristoph Hellwig status = __iomap_write_begin(iter, pos, len, page); 642afc51aaaSDarrick J. Wong 643afc51aaaSDarrick J. Wong if (unlikely(status)) 644afc51aaaSDarrick J. Wong goto out_unlock; 645afc51aaaSDarrick J. Wong 646afc51aaaSDarrick J. Wong *pagep = page; 647afc51aaaSDarrick J. Wong return 0; 648afc51aaaSDarrick J. Wong 649afc51aaaSDarrick J. Wong out_unlock: 650afc51aaaSDarrick J. Wong unlock_page(page); 651afc51aaaSDarrick J. Wong put_page(page); 6521b5c1e36SChristoph Hellwig iomap_write_failed(iter->inode, pos, len); 653afc51aaaSDarrick J. Wong 654afc51aaaSDarrick J. Wong out_no_page: 655afc51aaaSDarrick J. Wong if (page_ops && page_ops->page_done) 6561b5c1e36SChristoph Hellwig page_ops->page_done(iter->inode, pos, 0, NULL); 657afc51aaaSDarrick J. Wong return status; 658afc51aaaSDarrick J. Wong } 659afc51aaaSDarrick J. Wong 660e25ba8cbSMatthew Wilcox (Oracle) static size_t __iomap_write_end(struct inode *inode, loff_t pos, size_t len, 661e25ba8cbSMatthew Wilcox (Oracle) size_t copied, struct page *page) 662afc51aaaSDarrick J. Wong { 663cd1e5afeSMatthew Wilcox (Oracle) struct folio *folio = page_folio(page); 664cd1e5afeSMatthew Wilcox (Oracle) struct iomap_page *iop = to_iomap_page(folio); 665afc51aaaSDarrick J. Wong flush_dcache_page(page); 666afc51aaaSDarrick J. Wong 667afc51aaaSDarrick J. Wong /* 668afc51aaaSDarrick J. Wong * The blocks that were entirely written will now be uptodate, so we 669afc51aaaSDarrick J. Wong * don't have to worry about a readpage reading them and overwriting a 670f1f264b4SAndreas Gruenbacher * partial write. However, if we've encountered a short write and only 671afc51aaaSDarrick J. Wong * partially written into a block, it will not be marked uptodate, so a 672afc51aaaSDarrick J. Wong * readpage might come in and destroy our partial write. 673afc51aaaSDarrick J. Wong * 674f1f264b4SAndreas Gruenbacher * Do the simplest thing and just treat any short write to a 675f1f264b4SAndreas Gruenbacher * non-uptodate page as a zero-length write, and force the caller to 676f1f264b4SAndreas Gruenbacher * redo the whole thing. 677afc51aaaSDarrick J. Wong */ 678afc51aaaSDarrick J. Wong if (unlikely(copied < len && !PageUptodate(page))) 679afc51aaaSDarrick J. Wong return 0; 680*431c0566SMatthew Wilcox (Oracle) iomap_set_range_uptodate(folio, iop, offset_in_folio(folio, pos), len); 681fd7353f8SMatthew Wilcox (Oracle) __set_page_dirty_nobuffers(page); 682afc51aaaSDarrick J. Wong return copied; 683afc51aaaSDarrick J. Wong } 684afc51aaaSDarrick J. Wong 685fad0a1abSChristoph Hellwig static size_t iomap_write_end_inline(const struct iomap_iter *iter, 686fad0a1abSChristoph Hellwig struct page *page, loff_t pos, size_t copied) 687afc51aaaSDarrick J. Wong { 688fad0a1abSChristoph Hellwig const struct iomap *iomap = &iter->iomap; 689afc51aaaSDarrick J. Wong void *addr; 690afc51aaaSDarrick J. Wong 691afc51aaaSDarrick J. Wong WARN_ON_ONCE(!PageUptodate(page)); 69269f4a26cSGao Xiang BUG_ON(!iomap_inline_data_valid(iomap)); 693afc51aaaSDarrick J. Wong 6947ed3cd1aSMatthew Wilcox (Oracle) flush_dcache_page(page); 695ab069d5fSMatthew Wilcox (Oracle) addr = kmap_local_page(page) + pos; 696ab069d5fSMatthew Wilcox (Oracle) memcpy(iomap_inline_data(iomap, pos), addr, copied); 697ab069d5fSMatthew Wilcox (Oracle) kunmap_local(addr); 698afc51aaaSDarrick J. Wong 6991b5c1e36SChristoph Hellwig mark_inode_dirty(iter->inode); 700afc51aaaSDarrick J. Wong return copied; 701afc51aaaSDarrick J. Wong } 702afc51aaaSDarrick J. Wong 703e25ba8cbSMatthew Wilcox (Oracle) /* Returns the number of bytes copied. May be 0. Cannot be an errno. */ 7041b5c1e36SChristoph Hellwig static size_t iomap_write_end(struct iomap_iter *iter, loff_t pos, size_t len, 7051b5c1e36SChristoph Hellwig size_t copied, struct page *page) 706afc51aaaSDarrick J. Wong { 7071b5c1e36SChristoph Hellwig const struct iomap_page_ops *page_ops = iter->iomap.page_ops; 708fad0a1abSChristoph Hellwig const struct iomap *srcmap = iomap_iter_srcmap(iter); 7091b5c1e36SChristoph Hellwig loff_t old_size = iter->inode->i_size; 710e25ba8cbSMatthew Wilcox (Oracle) size_t ret; 711afc51aaaSDarrick J. Wong 712c039b997SGoldwyn Rodrigues if (srcmap->type == IOMAP_INLINE) { 7131b5c1e36SChristoph Hellwig ret = iomap_write_end_inline(iter, page, pos, copied); 714c039b997SGoldwyn Rodrigues } else if (srcmap->flags & IOMAP_F_BUFFER_HEAD) { 7151b5c1e36SChristoph Hellwig ret = block_write_end(NULL, iter->inode->i_mapping, pos, len, 7161b5c1e36SChristoph Hellwig copied, page, NULL); 717afc51aaaSDarrick J. Wong } else { 7181b5c1e36SChristoph Hellwig ret = __iomap_write_end(iter->inode, pos, len, copied, page); 719afc51aaaSDarrick J. Wong } 720afc51aaaSDarrick J. Wong 721afc51aaaSDarrick J. Wong /* 722afc51aaaSDarrick J. Wong * Update the in-memory inode size after copying the data into the page 723afc51aaaSDarrick J. Wong * cache. It's up to the file system to write the updated size to disk, 724afc51aaaSDarrick J. Wong * preferably after I/O completion so that no stale data is exposed. 725afc51aaaSDarrick J. Wong */ 726afc51aaaSDarrick J. Wong if (pos + ret > old_size) { 7271b5c1e36SChristoph Hellwig i_size_write(iter->inode, pos + ret); 7281b5c1e36SChristoph Hellwig iter->iomap.flags |= IOMAP_F_SIZE_CHANGED; 729afc51aaaSDarrick J. Wong } 730afc51aaaSDarrick J. Wong unlock_page(page); 731afc51aaaSDarrick J. Wong 732afc51aaaSDarrick J. Wong if (old_size < pos) 7331b5c1e36SChristoph Hellwig pagecache_isize_extended(iter->inode, old_size, pos); 734afc51aaaSDarrick J. Wong if (page_ops && page_ops->page_done) 7351b5c1e36SChristoph Hellwig page_ops->page_done(iter->inode, pos, ret, page); 736afc51aaaSDarrick J. Wong put_page(page); 737afc51aaaSDarrick J. Wong 738afc51aaaSDarrick J. Wong if (ret < len) 7391b5c1e36SChristoph Hellwig iomap_write_failed(iter->inode, pos, len); 740afc51aaaSDarrick J. Wong return ret; 741afc51aaaSDarrick J. Wong } 742afc51aaaSDarrick J. Wong 743ce83a025SChristoph Hellwig static loff_t iomap_write_iter(struct iomap_iter *iter, struct iov_iter *i) 744afc51aaaSDarrick J. Wong { 745ce83a025SChristoph Hellwig loff_t length = iomap_length(iter); 746ce83a025SChristoph Hellwig loff_t pos = iter->pos; 747afc51aaaSDarrick J. Wong ssize_t written = 0; 748ce83a025SChristoph Hellwig long status = 0; 749afc51aaaSDarrick J. Wong 750afc51aaaSDarrick J. Wong do { 751afc51aaaSDarrick J. Wong struct page *page; 752afc51aaaSDarrick J. Wong unsigned long offset; /* Offset into pagecache page */ 753afc51aaaSDarrick J. Wong unsigned long bytes; /* Bytes to write to page */ 754afc51aaaSDarrick J. Wong size_t copied; /* Bytes copied from user */ 755afc51aaaSDarrick J. Wong 756afc51aaaSDarrick J. Wong offset = offset_in_page(pos); 757afc51aaaSDarrick J. Wong bytes = min_t(unsigned long, PAGE_SIZE - offset, 758afc51aaaSDarrick J. Wong iov_iter_count(i)); 759afc51aaaSDarrick J. Wong again: 760afc51aaaSDarrick J. Wong if (bytes > length) 761afc51aaaSDarrick J. Wong bytes = length; 762afc51aaaSDarrick J. Wong 763afc51aaaSDarrick J. Wong /* 764f1f264b4SAndreas Gruenbacher * Bring in the user page that we'll copy from _first_. 765afc51aaaSDarrick J. Wong * Otherwise there's a nasty deadlock on copying from the 766afc51aaaSDarrick J. Wong * same page as we're writing to, without it being marked 767afc51aaaSDarrick J. Wong * up-to-date. 768afc51aaaSDarrick J. Wong */ 769a6294593SAndreas Gruenbacher if (unlikely(fault_in_iov_iter_readable(i, bytes))) { 770afc51aaaSDarrick J. Wong status = -EFAULT; 771afc51aaaSDarrick J. Wong break; 772afc51aaaSDarrick J. Wong } 773afc51aaaSDarrick J. Wong 774b74b1293SChristoph Hellwig status = iomap_write_begin(iter, pos, bytes, &page); 775afc51aaaSDarrick J. Wong if (unlikely(status)) 776afc51aaaSDarrick J. Wong break; 777afc51aaaSDarrick J. Wong 778ce83a025SChristoph Hellwig if (mapping_writably_mapped(iter->inode->i_mapping)) 779afc51aaaSDarrick J. Wong flush_dcache_page(page); 780afc51aaaSDarrick J. Wong 781f0b65f39SAl Viro copied = copy_page_from_iter_atomic(page, offset, bytes, i); 782afc51aaaSDarrick J. Wong 7831b5c1e36SChristoph Hellwig status = iomap_write_end(iter, pos, bytes, copied, page); 784afc51aaaSDarrick J. Wong 785f0b65f39SAl Viro if (unlikely(copied != status)) 786f0b65f39SAl Viro iov_iter_revert(i, copied - status); 787afc51aaaSDarrick J. Wong 788f0b65f39SAl Viro cond_resched(); 789bc1bb416SAl Viro if (unlikely(status == 0)) { 790afc51aaaSDarrick J. Wong /* 791bc1bb416SAl Viro * A short copy made iomap_write_end() reject the 792bc1bb416SAl Viro * thing entirely. Might be memory poisoning 793bc1bb416SAl Viro * halfway through, might be a race with munmap, 794bc1bb416SAl Viro * might be severe memory pressure. 795afc51aaaSDarrick J. Wong */ 796bc1bb416SAl Viro if (copied) 797bc1bb416SAl Viro bytes = copied; 798afc51aaaSDarrick J. Wong goto again; 799afc51aaaSDarrick J. Wong } 800f0b65f39SAl Viro pos += status; 801f0b65f39SAl Viro written += status; 802f0b65f39SAl Viro length -= status; 803afc51aaaSDarrick J. Wong 804ce83a025SChristoph Hellwig balance_dirty_pages_ratelimited(iter->inode->i_mapping); 805afc51aaaSDarrick J. Wong } while (iov_iter_count(i) && length); 806afc51aaaSDarrick J. Wong 807afc51aaaSDarrick J. Wong return written ? written : status; 808afc51aaaSDarrick J. Wong } 809afc51aaaSDarrick J. Wong 810afc51aaaSDarrick J. Wong ssize_t 811ce83a025SChristoph Hellwig iomap_file_buffered_write(struct kiocb *iocb, struct iov_iter *i, 812afc51aaaSDarrick J. Wong const struct iomap_ops *ops) 813afc51aaaSDarrick J. Wong { 814ce83a025SChristoph Hellwig struct iomap_iter iter = { 815ce83a025SChristoph Hellwig .inode = iocb->ki_filp->f_mapping->host, 816ce83a025SChristoph Hellwig .pos = iocb->ki_pos, 817ce83a025SChristoph Hellwig .len = iov_iter_count(i), 818ce83a025SChristoph Hellwig .flags = IOMAP_WRITE, 819ce83a025SChristoph Hellwig }; 820ce83a025SChristoph Hellwig int ret; 821afc51aaaSDarrick J. Wong 822ce83a025SChristoph Hellwig while ((ret = iomap_iter(&iter, ops)) > 0) 823ce83a025SChristoph Hellwig iter.processed = iomap_write_iter(&iter, i); 824ce83a025SChristoph Hellwig if (iter.pos == iocb->ki_pos) 825ce83a025SChristoph Hellwig return ret; 826ce83a025SChristoph Hellwig return iter.pos - iocb->ki_pos; 827afc51aaaSDarrick J. Wong } 828afc51aaaSDarrick J. Wong EXPORT_SYMBOL_GPL(iomap_file_buffered_write); 829afc51aaaSDarrick J. Wong 8308fc274d1SChristoph Hellwig static loff_t iomap_unshare_iter(struct iomap_iter *iter) 831afc51aaaSDarrick J. Wong { 8328fc274d1SChristoph Hellwig struct iomap *iomap = &iter->iomap; 833fad0a1abSChristoph Hellwig const struct iomap *srcmap = iomap_iter_srcmap(iter); 8348fc274d1SChristoph Hellwig loff_t pos = iter->pos; 8358fc274d1SChristoph Hellwig loff_t length = iomap_length(iter); 836afc51aaaSDarrick J. Wong long status = 0; 837d4ff3b2eSMatthew Wilcox (Oracle) loff_t written = 0; 838afc51aaaSDarrick J. Wong 8393590c4d8SChristoph Hellwig /* don't bother with blocks that are not shared to start with */ 8403590c4d8SChristoph Hellwig if (!(iomap->flags & IOMAP_F_SHARED)) 8413590c4d8SChristoph Hellwig return length; 8423590c4d8SChristoph Hellwig /* don't bother with holes or unwritten extents */ 843c039b997SGoldwyn Rodrigues if (srcmap->type == IOMAP_HOLE || srcmap->type == IOMAP_UNWRITTEN) 8443590c4d8SChristoph Hellwig return length; 8453590c4d8SChristoph Hellwig 846afc51aaaSDarrick J. Wong do { 84732a38a49SChristoph Hellwig unsigned long offset = offset_in_page(pos); 84832a38a49SChristoph Hellwig unsigned long bytes = min_t(loff_t, PAGE_SIZE - offset, length); 84932a38a49SChristoph Hellwig struct page *page; 850afc51aaaSDarrick J. Wong 851b74b1293SChristoph Hellwig status = iomap_write_begin(iter, pos, bytes, &page); 852afc51aaaSDarrick J. Wong if (unlikely(status)) 853afc51aaaSDarrick J. Wong return status; 854afc51aaaSDarrick J. Wong 8551b5c1e36SChristoph Hellwig status = iomap_write_end(iter, pos, bytes, bytes, page); 856afc51aaaSDarrick J. Wong if (WARN_ON_ONCE(status == 0)) 857afc51aaaSDarrick J. Wong return -EIO; 858afc51aaaSDarrick J. Wong 859afc51aaaSDarrick J. Wong cond_resched(); 860afc51aaaSDarrick J. Wong 861afc51aaaSDarrick J. Wong pos += status; 862afc51aaaSDarrick J. Wong written += status; 863afc51aaaSDarrick J. Wong length -= status; 864afc51aaaSDarrick J. Wong 8658fc274d1SChristoph Hellwig balance_dirty_pages_ratelimited(iter->inode->i_mapping); 866afc51aaaSDarrick J. Wong } while (length); 867afc51aaaSDarrick J. Wong 868afc51aaaSDarrick J. Wong return written; 869afc51aaaSDarrick J. Wong } 870afc51aaaSDarrick J. Wong 871afc51aaaSDarrick J. Wong int 8723590c4d8SChristoph Hellwig iomap_file_unshare(struct inode *inode, loff_t pos, loff_t len, 873afc51aaaSDarrick J. Wong const struct iomap_ops *ops) 874afc51aaaSDarrick J. Wong { 8758fc274d1SChristoph Hellwig struct iomap_iter iter = { 8768fc274d1SChristoph Hellwig .inode = inode, 8778fc274d1SChristoph Hellwig .pos = pos, 8788fc274d1SChristoph Hellwig .len = len, 879b74b1293SChristoph Hellwig .flags = IOMAP_WRITE | IOMAP_UNSHARE, 8808fc274d1SChristoph Hellwig }; 8818fc274d1SChristoph Hellwig int ret; 882afc51aaaSDarrick J. Wong 8838fc274d1SChristoph Hellwig while ((ret = iomap_iter(&iter, ops)) > 0) 8848fc274d1SChristoph Hellwig iter.processed = iomap_unshare_iter(&iter); 885afc51aaaSDarrick J. Wong return ret; 886afc51aaaSDarrick J. Wong } 8873590c4d8SChristoph Hellwig EXPORT_SYMBOL_GPL(iomap_file_unshare); 888afc51aaaSDarrick J. Wong 8891b5c1e36SChristoph Hellwig static s64 __iomap_zero_iter(struct iomap_iter *iter, loff_t pos, u64 length) 890afc51aaaSDarrick J. Wong { 891afc51aaaSDarrick J. Wong struct page *page; 892afc51aaaSDarrick J. Wong int status; 89381ee8e52SMatthew Wilcox (Oracle) unsigned offset = offset_in_page(pos); 89481ee8e52SMatthew Wilcox (Oracle) unsigned bytes = min_t(u64, PAGE_SIZE - offset, length); 895afc51aaaSDarrick J. Wong 896b74b1293SChristoph Hellwig status = iomap_write_begin(iter, pos, bytes, &page); 897afc51aaaSDarrick J. Wong if (status) 898afc51aaaSDarrick J. Wong return status; 899afc51aaaSDarrick J. Wong 900afc51aaaSDarrick J. Wong zero_user(page, offset, bytes); 901afc51aaaSDarrick J. Wong mark_page_accessed(page); 902afc51aaaSDarrick J. Wong 9031b5c1e36SChristoph Hellwig return iomap_write_end(iter, pos, bytes, bytes, page); 904afc51aaaSDarrick J. Wong } 905afc51aaaSDarrick J. Wong 9062aa3048eSChristoph Hellwig static loff_t iomap_zero_iter(struct iomap_iter *iter, bool *did_zero) 907afc51aaaSDarrick J. Wong { 9082aa3048eSChristoph Hellwig struct iomap *iomap = &iter->iomap; 909fad0a1abSChristoph Hellwig const struct iomap *srcmap = iomap_iter_srcmap(iter); 9102aa3048eSChristoph Hellwig loff_t pos = iter->pos; 9112aa3048eSChristoph Hellwig loff_t length = iomap_length(iter); 912afc51aaaSDarrick J. Wong loff_t written = 0; 913afc51aaaSDarrick J. Wong 914afc51aaaSDarrick J. Wong /* already zeroed? we're done. */ 915c039b997SGoldwyn Rodrigues if (srcmap->type == IOMAP_HOLE || srcmap->type == IOMAP_UNWRITTEN) 91681ee8e52SMatthew Wilcox (Oracle) return length; 917afc51aaaSDarrick J. Wong 918afc51aaaSDarrick J. Wong do { 91981ee8e52SMatthew Wilcox (Oracle) s64 bytes; 920afc51aaaSDarrick J. Wong 9212aa3048eSChristoph Hellwig if (IS_DAX(iter->inode)) 92281ee8e52SMatthew Wilcox (Oracle) bytes = dax_iomap_zero(pos, length, iomap); 923afc51aaaSDarrick J. Wong else 9241b5c1e36SChristoph Hellwig bytes = __iomap_zero_iter(iter, pos, length); 92581ee8e52SMatthew Wilcox (Oracle) if (bytes < 0) 92681ee8e52SMatthew Wilcox (Oracle) return bytes; 927afc51aaaSDarrick J. Wong 928afc51aaaSDarrick J. Wong pos += bytes; 92981ee8e52SMatthew Wilcox (Oracle) length -= bytes; 930afc51aaaSDarrick J. Wong written += bytes; 931afc51aaaSDarrick J. Wong if (did_zero) 932afc51aaaSDarrick J. Wong *did_zero = true; 93381ee8e52SMatthew Wilcox (Oracle) } while (length > 0); 934afc51aaaSDarrick J. Wong 935afc51aaaSDarrick J. Wong return written; 936afc51aaaSDarrick J. Wong } 937afc51aaaSDarrick J. Wong 938afc51aaaSDarrick J. Wong int 939afc51aaaSDarrick J. Wong iomap_zero_range(struct inode *inode, loff_t pos, loff_t len, bool *did_zero, 940afc51aaaSDarrick J. Wong const struct iomap_ops *ops) 941afc51aaaSDarrick J. Wong { 9422aa3048eSChristoph Hellwig struct iomap_iter iter = { 9432aa3048eSChristoph Hellwig .inode = inode, 9442aa3048eSChristoph Hellwig .pos = pos, 9452aa3048eSChristoph Hellwig .len = len, 9462aa3048eSChristoph Hellwig .flags = IOMAP_ZERO, 9472aa3048eSChristoph Hellwig }; 9482aa3048eSChristoph Hellwig int ret; 949afc51aaaSDarrick J. Wong 9502aa3048eSChristoph Hellwig while ((ret = iomap_iter(&iter, ops)) > 0) 9512aa3048eSChristoph Hellwig iter.processed = iomap_zero_iter(&iter, did_zero); 952afc51aaaSDarrick J. Wong return ret; 953afc51aaaSDarrick J. Wong } 954afc51aaaSDarrick J. Wong EXPORT_SYMBOL_GPL(iomap_zero_range); 955afc51aaaSDarrick J. Wong 956afc51aaaSDarrick J. Wong int 957afc51aaaSDarrick J. Wong iomap_truncate_page(struct inode *inode, loff_t pos, bool *did_zero, 958afc51aaaSDarrick J. Wong const struct iomap_ops *ops) 959afc51aaaSDarrick J. Wong { 960afc51aaaSDarrick J. Wong unsigned int blocksize = i_blocksize(inode); 961afc51aaaSDarrick J. Wong unsigned int off = pos & (blocksize - 1); 962afc51aaaSDarrick J. Wong 963afc51aaaSDarrick J. Wong /* Block boundary? Nothing to do */ 964afc51aaaSDarrick J. Wong if (!off) 965afc51aaaSDarrick J. Wong return 0; 966afc51aaaSDarrick J. Wong return iomap_zero_range(inode, pos, blocksize - off, did_zero, ops); 967afc51aaaSDarrick J. Wong } 968afc51aaaSDarrick J. Wong EXPORT_SYMBOL_GPL(iomap_truncate_page); 969afc51aaaSDarrick J. Wong 970253564baSChristoph Hellwig static loff_t iomap_page_mkwrite_iter(struct iomap_iter *iter, 971253564baSChristoph Hellwig struct page *page) 972afc51aaaSDarrick J. Wong { 973d1bd0b4eSMatthew Wilcox (Oracle) struct folio *folio = page_folio(page); 974253564baSChristoph Hellwig loff_t length = iomap_length(iter); 975afc51aaaSDarrick J. Wong int ret; 976afc51aaaSDarrick J. Wong 977253564baSChristoph Hellwig if (iter->iomap.flags & IOMAP_F_BUFFER_HEAD) { 978d1bd0b4eSMatthew Wilcox (Oracle) ret = __block_write_begin_int(folio, iter->pos, length, NULL, 979253564baSChristoph Hellwig &iter->iomap); 980afc51aaaSDarrick J. Wong if (ret) 981afc51aaaSDarrick J. Wong return ret; 982afc51aaaSDarrick J. Wong block_commit_write(page, 0, length); 983afc51aaaSDarrick J. Wong } else { 984afc51aaaSDarrick J. Wong WARN_ON_ONCE(!PageUptodate(page)); 985afc51aaaSDarrick J. Wong set_page_dirty(page); 986afc51aaaSDarrick J. Wong } 987afc51aaaSDarrick J. Wong 988afc51aaaSDarrick J. Wong return length; 989afc51aaaSDarrick J. Wong } 990afc51aaaSDarrick J. Wong 991afc51aaaSDarrick J. Wong vm_fault_t iomap_page_mkwrite(struct vm_fault *vmf, const struct iomap_ops *ops) 992afc51aaaSDarrick J. Wong { 993253564baSChristoph Hellwig struct iomap_iter iter = { 994253564baSChristoph Hellwig .inode = file_inode(vmf->vma->vm_file), 995253564baSChristoph Hellwig .flags = IOMAP_WRITE | IOMAP_FAULT, 996253564baSChristoph Hellwig }; 997afc51aaaSDarrick J. Wong struct page *page = vmf->page; 998afc51aaaSDarrick J. Wong ssize_t ret; 999afc51aaaSDarrick J. Wong 1000afc51aaaSDarrick J. Wong lock_page(page); 1001253564baSChristoph Hellwig ret = page_mkwrite_check_truncate(page, iter.inode); 1002243145bcSAndreas Gruenbacher if (ret < 0) 1003afc51aaaSDarrick J. Wong goto out_unlock; 1004253564baSChristoph Hellwig iter.pos = page_offset(page); 1005253564baSChristoph Hellwig iter.len = ret; 1006253564baSChristoph Hellwig while ((ret = iomap_iter(&iter, ops)) > 0) 1007253564baSChristoph Hellwig iter.processed = iomap_page_mkwrite_iter(&iter, page); 1008afc51aaaSDarrick J. Wong 1009253564baSChristoph Hellwig if (ret < 0) 1010afc51aaaSDarrick J. Wong goto out_unlock; 1011afc51aaaSDarrick J. Wong wait_for_stable_page(page); 1012afc51aaaSDarrick J. Wong return VM_FAULT_LOCKED; 1013afc51aaaSDarrick J. Wong out_unlock: 1014afc51aaaSDarrick J. Wong unlock_page(page); 1015afc51aaaSDarrick J. Wong return block_page_mkwrite_return(ret); 1016afc51aaaSDarrick J. Wong } 1017afc51aaaSDarrick J. Wong EXPORT_SYMBOL_GPL(iomap_page_mkwrite); 1018598ecfbaSChristoph Hellwig 10198ffd74e9SMatthew Wilcox (Oracle) static void iomap_finish_folio_write(struct inode *inode, struct folio *folio, 10208ffd74e9SMatthew Wilcox (Oracle) size_t len, int error) 1021598ecfbaSChristoph Hellwig { 102295c4cd05SMatthew Wilcox (Oracle) struct iomap_page *iop = to_iomap_page(folio); 1023598ecfbaSChristoph Hellwig 1024598ecfbaSChristoph Hellwig if (error) { 10258ffd74e9SMatthew Wilcox (Oracle) folio_set_error(folio); 1026b69eea82SDarrick J. Wong mapping_set_error(inode->i_mapping, error); 1027598ecfbaSChristoph Hellwig } 1028598ecfbaSChristoph Hellwig 10298ffd74e9SMatthew Wilcox (Oracle) WARN_ON_ONCE(i_blocks_per_folio(inode, folio) > 1 && !iop); 10300fb2d720SMatthew Wilcox (Oracle) WARN_ON_ONCE(iop && atomic_read(&iop->write_bytes_pending) <= 0); 1031598ecfbaSChristoph Hellwig 10320fb2d720SMatthew Wilcox (Oracle) if (!iop || atomic_sub_and_test(len, &iop->write_bytes_pending)) 10338ffd74e9SMatthew Wilcox (Oracle) folio_end_writeback(folio); 1034598ecfbaSChristoph Hellwig } 1035598ecfbaSChristoph Hellwig 1036598ecfbaSChristoph Hellwig /* 1037598ecfbaSChristoph Hellwig * We're now finished for good with this ioend structure. Update the page 1038598ecfbaSChristoph Hellwig * state, release holds on bios, and finally free up memory. Do not use the 1039598ecfbaSChristoph Hellwig * ioend after this. 1040598ecfbaSChristoph Hellwig */ 1041598ecfbaSChristoph Hellwig static void 1042598ecfbaSChristoph Hellwig iomap_finish_ioend(struct iomap_ioend *ioend, int error) 1043598ecfbaSChristoph Hellwig { 1044598ecfbaSChristoph Hellwig struct inode *inode = ioend->io_inode; 1045598ecfbaSChristoph Hellwig struct bio *bio = &ioend->io_inline_bio; 1046598ecfbaSChristoph Hellwig struct bio *last = ioend->io_bio, *next; 1047598ecfbaSChristoph Hellwig u64 start = bio->bi_iter.bi_sector; 1048c275779fSZorro Lang loff_t offset = ioend->io_offset; 1049598ecfbaSChristoph Hellwig bool quiet = bio_flagged(bio, BIO_QUIET); 1050598ecfbaSChristoph Hellwig 1051598ecfbaSChristoph Hellwig for (bio = &ioend->io_inline_bio; bio; bio = next) { 10528ffd74e9SMatthew Wilcox (Oracle) struct folio_iter fi; 1053598ecfbaSChristoph Hellwig 1054598ecfbaSChristoph Hellwig /* 1055598ecfbaSChristoph Hellwig * For the last bio, bi_private points to the ioend, so we 1056598ecfbaSChristoph Hellwig * need to explicitly end the iteration here. 1057598ecfbaSChristoph Hellwig */ 1058598ecfbaSChristoph Hellwig if (bio == last) 1059598ecfbaSChristoph Hellwig next = NULL; 1060598ecfbaSChristoph Hellwig else 1061598ecfbaSChristoph Hellwig next = bio->bi_private; 1062598ecfbaSChristoph Hellwig 10638ffd74e9SMatthew Wilcox (Oracle) /* walk all folios in bio, ending page IO on them */ 10648ffd74e9SMatthew Wilcox (Oracle) bio_for_each_folio_all(fi, bio) 10658ffd74e9SMatthew Wilcox (Oracle) iomap_finish_folio_write(inode, fi.folio, fi.length, 10668ffd74e9SMatthew Wilcox (Oracle) error); 1067598ecfbaSChristoph Hellwig bio_put(bio); 1068598ecfbaSChristoph Hellwig } 1069c275779fSZorro Lang /* The ioend has been freed by bio_put() */ 1070598ecfbaSChristoph Hellwig 1071598ecfbaSChristoph Hellwig if (unlikely(error && !quiet)) { 1072598ecfbaSChristoph Hellwig printk_ratelimited(KERN_ERR 10739cd0ed63SDarrick J. Wong "%s: writeback error on inode %lu, offset %lld, sector %llu", 1074c275779fSZorro Lang inode->i_sb->s_id, inode->i_ino, offset, start); 1075598ecfbaSChristoph Hellwig } 1076598ecfbaSChristoph Hellwig } 1077598ecfbaSChristoph Hellwig 1078598ecfbaSChristoph Hellwig void 1079598ecfbaSChristoph Hellwig iomap_finish_ioends(struct iomap_ioend *ioend, int error) 1080598ecfbaSChristoph Hellwig { 1081598ecfbaSChristoph Hellwig struct list_head tmp; 1082598ecfbaSChristoph Hellwig 1083598ecfbaSChristoph Hellwig list_replace_init(&ioend->io_list, &tmp); 1084598ecfbaSChristoph Hellwig iomap_finish_ioend(ioend, error); 1085598ecfbaSChristoph Hellwig 1086598ecfbaSChristoph Hellwig while (!list_empty(&tmp)) { 1087598ecfbaSChristoph Hellwig ioend = list_first_entry(&tmp, struct iomap_ioend, io_list); 1088598ecfbaSChristoph Hellwig list_del_init(&ioend->io_list); 1089598ecfbaSChristoph Hellwig iomap_finish_ioend(ioend, error); 1090598ecfbaSChristoph Hellwig } 1091598ecfbaSChristoph Hellwig } 1092598ecfbaSChristoph Hellwig EXPORT_SYMBOL_GPL(iomap_finish_ioends); 1093598ecfbaSChristoph Hellwig 1094598ecfbaSChristoph Hellwig /* 1095598ecfbaSChristoph Hellwig * We can merge two adjacent ioends if they have the same set of work to do. 1096598ecfbaSChristoph Hellwig */ 1097598ecfbaSChristoph Hellwig static bool 1098598ecfbaSChristoph Hellwig iomap_ioend_can_merge(struct iomap_ioend *ioend, struct iomap_ioend *next) 1099598ecfbaSChristoph Hellwig { 1100598ecfbaSChristoph Hellwig if (ioend->io_bio->bi_status != next->io_bio->bi_status) 1101598ecfbaSChristoph Hellwig return false; 1102598ecfbaSChristoph Hellwig if ((ioend->io_flags & IOMAP_F_SHARED) ^ 1103598ecfbaSChristoph Hellwig (next->io_flags & IOMAP_F_SHARED)) 1104598ecfbaSChristoph Hellwig return false; 1105598ecfbaSChristoph Hellwig if ((ioend->io_type == IOMAP_UNWRITTEN) ^ 1106598ecfbaSChristoph Hellwig (next->io_type == IOMAP_UNWRITTEN)) 1107598ecfbaSChristoph Hellwig return false; 1108598ecfbaSChristoph Hellwig if (ioend->io_offset + ioend->io_size != next->io_offset) 1109598ecfbaSChristoph Hellwig return false; 1110598ecfbaSChristoph Hellwig return true; 1111598ecfbaSChristoph Hellwig } 1112598ecfbaSChristoph Hellwig 1113598ecfbaSChristoph Hellwig void 11146e552494SBrian Foster iomap_ioend_try_merge(struct iomap_ioend *ioend, struct list_head *more_ioends) 1115598ecfbaSChristoph Hellwig { 1116598ecfbaSChristoph Hellwig struct iomap_ioend *next; 1117598ecfbaSChristoph Hellwig 1118598ecfbaSChristoph Hellwig INIT_LIST_HEAD(&ioend->io_list); 1119598ecfbaSChristoph Hellwig 1120598ecfbaSChristoph Hellwig while ((next = list_first_entry_or_null(more_ioends, struct iomap_ioend, 1121598ecfbaSChristoph Hellwig io_list))) { 1122598ecfbaSChristoph Hellwig if (!iomap_ioend_can_merge(ioend, next)) 1123598ecfbaSChristoph Hellwig break; 1124598ecfbaSChristoph Hellwig list_move_tail(&next->io_list, &ioend->io_list); 1125598ecfbaSChristoph Hellwig ioend->io_size += next->io_size; 1126598ecfbaSChristoph Hellwig } 1127598ecfbaSChristoph Hellwig } 1128598ecfbaSChristoph Hellwig EXPORT_SYMBOL_GPL(iomap_ioend_try_merge); 1129598ecfbaSChristoph Hellwig 1130598ecfbaSChristoph Hellwig static int 11314f0f586bSSami Tolvanen iomap_ioend_compare(void *priv, const struct list_head *a, 11324f0f586bSSami Tolvanen const struct list_head *b) 1133598ecfbaSChristoph Hellwig { 1134b3d423ecSChristoph Hellwig struct iomap_ioend *ia = container_of(a, struct iomap_ioend, io_list); 1135b3d423ecSChristoph Hellwig struct iomap_ioend *ib = container_of(b, struct iomap_ioend, io_list); 1136598ecfbaSChristoph Hellwig 1137598ecfbaSChristoph Hellwig if (ia->io_offset < ib->io_offset) 1138598ecfbaSChristoph Hellwig return -1; 1139b3d423ecSChristoph Hellwig if (ia->io_offset > ib->io_offset) 1140598ecfbaSChristoph Hellwig return 1; 1141598ecfbaSChristoph Hellwig return 0; 1142598ecfbaSChristoph Hellwig } 1143598ecfbaSChristoph Hellwig 1144598ecfbaSChristoph Hellwig void 1145598ecfbaSChristoph Hellwig iomap_sort_ioends(struct list_head *ioend_list) 1146598ecfbaSChristoph Hellwig { 1147598ecfbaSChristoph Hellwig list_sort(NULL, ioend_list, iomap_ioend_compare); 1148598ecfbaSChristoph Hellwig } 1149598ecfbaSChristoph Hellwig EXPORT_SYMBOL_GPL(iomap_sort_ioends); 1150598ecfbaSChristoph Hellwig 1151598ecfbaSChristoph Hellwig static void iomap_writepage_end_bio(struct bio *bio) 1152598ecfbaSChristoph Hellwig { 1153598ecfbaSChristoph Hellwig struct iomap_ioend *ioend = bio->bi_private; 1154598ecfbaSChristoph Hellwig 1155598ecfbaSChristoph Hellwig iomap_finish_ioend(ioend, blk_status_to_errno(bio->bi_status)); 1156598ecfbaSChristoph Hellwig } 1157598ecfbaSChristoph Hellwig 1158598ecfbaSChristoph Hellwig /* 1159598ecfbaSChristoph Hellwig * Submit the final bio for an ioend. 1160598ecfbaSChristoph Hellwig * 1161598ecfbaSChristoph Hellwig * If @error is non-zero, it means that we have a situation where some part of 1162f1f264b4SAndreas Gruenbacher * the submission process has failed after we've marked pages for writeback 1163598ecfbaSChristoph Hellwig * and unlocked them. In this situation, we need to fail the bio instead of 1164598ecfbaSChristoph Hellwig * submitting it. This typically only happens on a filesystem shutdown. 1165598ecfbaSChristoph Hellwig */ 1166598ecfbaSChristoph Hellwig static int 1167598ecfbaSChristoph Hellwig iomap_submit_ioend(struct iomap_writepage_ctx *wpc, struct iomap_ioend *ioend, 1168598ecfbaSChristoph Hellwig int error) 1169598ecfbaSChristoph Hellwig { 1170598ecfbaSChristoph Hellwig ioend->io_bio->bi_private = ioend; 1171598ecfbaSChristoph Hellwig ioend->io_bio->bi_end_io = iomap_writepage_end_bio; 1172598ecfbaSChristoph Hellwig 1173598ecfbaSChristoph Hellwig if (wpc->ops->prepare_ioend) 1174598ecfbaSChristoph Hellwig error = wpc->ops->prepare_ioend(ioend, error); 1175598ecfbaSChristoph Hellwig if (error) { 1176598ecfbaSChristoph Hellwig /* 1177f1f264b4SAndreas Gruenbacher * If we're failing the IO now, just mark the ioend with an 1178598ecfbaSChristoph Hellwig * error and finish it. This will run IO completion immediately 1179598ecfbaSChristoph Hellwig * as there is only one reference to the ioend at this point in 1180598ecfbaSChristoph Hellwig * time. 1181598ecfbaSChristoph Hellwig */ 1182598ecfbaSChristoph Hellwig ioend->io_bio->bi_status = errno_to_blk_status(error); 1183598ecfbaSChristoph Hellwig bio_endio(ioend->io_bio); 1184598ecfbaSChristoph Hellwig return error; 1185598ecfbaSChristoph Hellwig } 1186598ecfbaSChristoph Hellwig 1187598ecfbaSChristoph Hellwig submit_bio(ioend->io_bio); 1188598ecfbaSChristoph Hellwig return 0; 1189598ecfbaSChristoph Hellwig } 1190598ecfbaSChristoph Hellwig 1191598ecfbaSChristoph Hellwig static struct iomap_ioend * 1192598ecfbaSChristoph Hellwig iomap_alloc_ioend(struct inode *inode, struct iomap_writepage_ctx *wpc, 1193598ecfbaSChristoph Hellwig loff_t offset, sector_t sector, struct writeback_control *wbc) 1194598ecfbaSChristoph Hellwig { 1195598ecfbaSChristoph Hellwig struct iomap_ioend *ioend; 1196598ecfbaSChristoph Hellwig struct bio *bio; 1197598ecfbaSChristoph Hellwig 1198a8affc03SChristoph Hellwig bio = bio_alloc_bioset(GFP_NOFS, BIO_MAX_VECS, &iomap_ioend_bioset); 1199598ecfbaSChristoph Hellwig bio_set_dev(bio, wpc->iomap.bdev); 1200598ecfbaSChristoph Hellwig bio->bi_iter.bi_sector = sector; 1201598ecfbaSChristoph Hellwig bio->bi_opf = REQ_OP_WRITE | wbc_to_write_flags(wbc); 1202598ecfbaSChristoph Hellwig bio->bi_write_hint = inode->i_write_hint; 1203598ecfbaSChristoph Hellwig wbc_init_bio(wbc, bio); 1204598ecfbaSChristoph Hellwig 1205598ecfbaSChristoph Hellwig ioend = container_of(bio, struct iomap_ioend, io_inline_bio); 1206598ecfbaSChristoph Hellwig INIT_LIST_HEAD(&ioend->io_list); 1207598ecfbaSChristoph Hellwig ioend->io_type = wpc->iomap.type; 1208598ecfbaSChristoph Hellwig ioend->io_flags = wpc->iomap.flags; 1209598ecfbaSChristoph Hellwig ioend->io_inode = inode; 1210598ecfbaSChristoph Hellwig ioend->io_size = 0; 1211598ecfbaSChristoph Hellwig ioend->io_offset = offset; 1212598ecfbaSChristoph Hellwig ioend->io_bio = bio; 1213598ecfbaSChristoph Hellwig return ioend; 1214598ecfbaSChristoph Hellwig } 1215598ecfbaSChristoph Hellwig 1216598ecfbaSChristoph Hellwig /* 1217598ecfbaSChristoph Hellwig * Allocate a new bio, and chain the old bio to the new one. 1218598ecfbaSChristoph Hellwig * 1219f1f264b4SAndreas Gruenbacher * Note that we have to perform the chaining in this unintuitive order 1220598ecfbaSChristoph Hellwig * so that the bi_private linkage is set up in the right direction for the 1221598ecfbaSChristoph Hellwig * traversal in iomap_finish_ioend(). 1222598ecfbaSChristoph Hellwig */ 1223598ecfbaSChristoph Hellwig static struct bio * 1224598ecfbaSChristoph Hellwig iomap_chain_bio(struct bio *prev) 1225598ecfbaSChristoph Hellwig { 1226598ecfbaSChristoph Hellwig struct bio *new; 1227598ecfbaSChristoph Hellwig 1228a8affc03SChristoph Hellwig new = bio_alloc(GFP_NOFS, BIO_MAX_VECS); 1229598ecfbaSChristoph Hellwig bio_copy_dev(new, prev);/* also copies over blkcg information */ 1230598ecfbaSChristoph Hellwig new->bi_iter.bi_sector = bio_end_sector(prev); 1231598ecfbaSChristoph Hellwig new->bi_opf = prev->bi_opf; 1232598ecfbaSChristoph Hellwig new->bi_write_hint = prev->bi_write_hint; 1233598ecfbaSChristoph Hellwig 1234598ecfbaSChristoph Hellwig bio_chain(prev, new); 1235598ecfbaSChristoph Hellwig bio_get(prev); /* for iomap_finish_ioend */ 1236598ecfbaSChristoph Hellwig submit_bio(prev); 1237598ecfbaSChristoph Hellwig return new; 1238598ecfbaSChristoph Hellwig } 1239598ecfbaSChristoph Hellwig 1240598ecfbaSChristoph Hellwig static bool 1241598ecfbaSChristoph Hellwig iomap_can_add_to_ioend(struct iomap_writepage_ctx *wpc, loff_t offset, 1242598ecfbaSChristoph Hellwig sector_t sector) 1243598ecfbaSChristoph Hellwig { 1244598ecfbaSChristoph Hellwig if ((wpc->iomap.flags & IOMAP_F_SHARED) != 1245598ecfbaSChristoph Hellwig (wpc->ioend->io_flags & IOMAP_F_SHARED)) 1246598ecfbaSChristoph Hellwig return false; 1247598ecfbaSChristoph Hellwig if (wpc->iomap.type != wpc->ioend->io_type) 1248598ecfbaSChristoph Hellwig return false; 1249598ecfbaSChristoph Hellwig if (offset != wpc->ioend->io_offset + wpc->ioend->io_size) 1250598ecfbaSChristoph Hellwig return false; 1251598ecfbaSChristoph Hellwig if (sector != bio_end_sector(wpc->ioend->io_bio)) 1252598ecfbaSChristoph Hellwig return false; 1253598ecfbaSChristoph Hellwig return true; 1254598ecfbaSChristoph Hellwig } 1255598ecfbaSChristoph Hellwig 1256598ecfbaSChristoph Hellwig /* 1257598ecfbaSChristoph Hellwig * Test to see if we have an existing ioend structure that we could append to 1258f1f264b4SAndreas Gruenbacher * first; otherwise finish off the current ioend and start another. 1259598ecfbaSChristoph Hellwig */ 1260598ecfbaSChristoph Hellwig static void 1261598ecfbaSChristoph Hellwig iomap_add_to_ioend(struct inode *inode, loff_t offset, struct page *page, 1262598ecfbaSChristoph Hellwig struct iomap_page *iop, struct iomap_writepage_ctx *wpc, 1263598ecfbaSChristoph Hellwig struct writeback_control *wbc, struct list_head *iolist) 1264598ecfbaSChristoph Hellwig { 1265598ecfbaSChristoph Hellwig sector_t sector = iomap_sector(&wpc->iomap, offset); 1266598ecfbaSChristoph Hellwig unsigned len = i_blocksize(inode); 1267598ecfbaSChristoph Hellwig unsigned poff = offset & (PAGE_SIZE - 1); 1268598ecfbaSChristoph Hellwig 1269598ecfbaSChristoph Hellwig if (!wpc->ioend || !iomap_can_add_to_ioend(wpc, offset, sector)) { 1270598ecfbaSChristoph Hellwig if (wpc->ioend) 1271598ecfbaSChristoph Hellwig list_add(&wpc->ioend->io_list, iolist); 1272598ecfbaSChristoph Hellwig wpc->ioend = iomap_alloc_ioend(inode, wpc, offset, sector, wbc); 1273598ecfbaSChristoph Hellwig } 1274598ecfbaSChristoph Hellwig 1275c1b79f11SChristoph Hellwig if (bio_add_page(wpc->ioend->io_bio, page, len, poff) != len) { 1276c1b79f11SChristoph Hellwig wpc->ioend->io_bio = iomap_chain_bio(wpc->ioend->io_bio); 1277c1b79f11SChristoph Hellwig __bio_add_page(wpc->ioend->io_bio, page, len, poff); 1278c1b79f11SChristoph Hellwig } 1279c1b79f11SChristoph Hellwig 12800fb2d720SMatthew Wilcox (Oracle) if (iop) 12810fb2d720SMatthew Wilcox (Oracle) atomic_add(len, &iop->write_bytes_pending); 1282598ecfbaSChristoph Hellwig wpc->ioend->io_size += len; 1283598ecfbaSChristoph Hellwig wbc_account_cgroup_owner(wbc, page, len); 1284598ecfbaSChristoph Hellwig } 1285598ecfbaSChristoph Hellwig 1286598ecfbaSChristoph Hellwig /* 1287598ecfbaSChristoph Hellwig * We implement an immediate ioend submission policy here to avoid needing to 1288598ecfbaSChristoph Hellwig * chain multiple ioends and hence nest mempool allocations which can violate 1289f1f264b4SAndreas Gruenbacher * the forward progress guarantees we need to provide. The current ioend we're 1290f1f264b4SAndreas Gruenbacher * adding blocks to is cached in the writepage context, and if the new block 1291f1f264b4SAndreas Gruenbacher * doesn't append to the cached ioend, it will create a new ioend and cache that 1292598ecfbaSChristoph Hellwig * instead. 1293598ecfbaSChristoph Hellwig * 1294598ecfbaSChristoph Hellwig * If a new ioend is created and cached, the old ioend is returned and queued 1295598ecfbaSChristoph Hellwig * locally for submission once the entire page is processed or an error has been 1296598ecfbaSChristoph Hellwig * detected. While ioends are submitted immediately after they are completed, 1297598ecfbaSChristoph Hellwig * batching optimisations are provided by higher level block plugging. 1298598ecfbaSChristoph Hellwig * 1299598ecfbaSChristoph Hellwig * At the end of a writeback pass, there will be a cached ioend remaining on the 1300598ecfbaSChristoph Hellwig * writepage context that the caller will need to submit. 1301598ecfbaSChristoph Hellwig */ 1302598ecfbaSChristoph Hellwig static int 1303598ecfbaSChristoph Hellwig iomap_writepage_map(struct iomap_writepage_ctx *wpc, 1304598ecfbaSChristoph Hellwig struct writeback_control *wbc, struct inode *inode, 1305598ecfbaSChristoph Hellwig struct page *page, u64 end_offset) 1306598ecfbaSChristoph Hellwig { 1307435d44b3SMatthew Wilcox (Oracle) struct folio *folio = page_folio(page); 1308435d44b3SMatthew Wilcox (Oracle) struct iomap_page *iop = iomap_page_create(inode, folio); 1309598ecfbaSChristoph Hellwig struct iomap_ioend *ioend, *next; 1310598ecfbaSChristoph Hellwig unsigned len = i_blocksize(inode); 1311598ecfbaSChristoph Hellwig u64 file_offset; /* file offset of page */ 1312598ecfbaSChristoph Hellwig int error = 0, count = 0, i; 1313598ecfbaSChristoph Hellwig LIST_HEAD(submit_list); 1314598ecfbaSChristoph Hellwig 13150fb2d720SMatthew Wilcox (Oracle) WARN_ON_ONCE(iop && atomic_read(&iop->write_bytes_pending) != 0); 1316598ecfbaSChristoph Hellwig 1317598ecfbaSChristoph Hellwig /* 1318598ecfbaSChristoph Hellwig * Walk through the page to find areas to write back. If we run off the 1319598ecfbaSChristoph Hellwig * end of the current map or find the current map invalid, grab a new 1320598ecfbaSChristoph Hellwig * one. 1321598ecfbaSChristoph Hellwig */ 1322598ecfbaSChristoph Hellwig for (i = 0, file_offset = page_offset(page); 1323598ecfbaSChristoph Hellwig i < (PAGE_SIZE >> inode->i_blkbits) && file_offset < end_offset; 1324598ecfbaSChristoph Hellwig i++, file_offset += len) { 1325598ecfbaSChristoph Hellwig if (iop && !test_bit(i, iop->uptodate)) 1326598ecfbaSChristoph Hellwig continue; 1327598ecfbaSChristoph Hellwig 1328598ecfbaSChristoph Hellwig error = wpc->ops->map_blocks(wpc, inode, file_offset); 1329598ecfbaSChristoph Hellwig if (error) 1330598ecfbaSChristoph Hellwig break; 13313e19e6f3SChristoph Hellwig if (WARN_ON_ONCE(wpc->iomap.type == IOMAP_INLINE)) 13323e19e6f3SChristoph Hellwig continue; 1333598ecfbaSChristoph Hellwig if (wpc->iomap.type == IOMAP_HOLE) 1334598ecfbaSChristoph Hellwig continue; 1335598ecfbaSChristoph Hellwig iomap_add_to_ioend(inode, file_offset, page, iop, wpc, wbc, 1336598ecfbaSChristoph Hellwig &submit_list); 1337598ecfbaSChristoph Hellwig count++; 1338598ecfbaSChristoph Hellwig } 1339598ecfbaSChristoph Hellwig 1340598ecfbaSChristoph Hellwig WARN_ON_ONCE(!wpc->ioend && !list_empty(&submit_list)); 1341598ecfbaSChristoph Hellwig WARN_ON_ONCE(!PageLocked(page)); 1342598ecfbaSChristoph Hellwig WARN_ON_ONCE(PageWriteback(page)); 134350e7d6c7SBrian Foster WARN_ON_ONCE(PageDirty(page)); 1344598ecfbaSChristoph Hellwig 1345598ecfbaSChristoph Hellwig /* 1346598ecfbaSChristoph Hellwig * We cannot cancel the ioend directly here on error. We may have 1347598ecfbaSChristoph Hellwig * already set other pages under writeback and hence we have to run I/O 1348598ecfbaSChristoph Hellwig * completion to mark the error state of the pages under writeback 1349598ecfbaSChristoph Hellwig * appropriately. 1350598ecfbaSChristoph Hellwig */ 1351598ecfbaSChristoph Hellwig if (unlikely(error)) { 1352598ecfbaSChristoph Hellwig /* 1353763e4cdcSBrian Foster * Let the filesystem know what portion of the current page 1354f1f264b4SAndreas Gruenbacher * failed to map. If the page hasn't been added to ioend, it 1355763e4cdcSBrian Foster * won't be affected by I/O completion and we must unlock it 1356763e4cdcSBrian Foster * now. 1357598ecfbaSChristoph Hellwig */ 1358598ecfbaSChristoph Hellwig if (wpc->ops->discard_page) 1359763e4cdcSBrian Foster wpc->ops->discard_page(page, file_offset); 1360763e4cdcSBrian Foster if (!count) { 1361598ecfbaSChristoph Hellwig ClearPageUptodate(page); 1362598ecfbaSChristoph Hellwig unlock_page(page); 1363598ecfbaSChristoph Hellwig goto done; 1364598ecfbaSChristoph Hellwig } 1365598ecfbaSChristoph Hellwig } 1366598ecfbaSChristoph Hellwig 136750e7d6c7SBrian Foster set_page_writeback(page); 1368598ecfbaSChristoph Hellwig unlock_page(page); 1369598ecfbaSChristoph Hellwig 1370598ecfbaSChristoph Hellwig /* 1371f1f264b4SAndreas Gruenbacher * Preserve the original error if there was one; catch 1372598ecfbaSChristoph Hellwig * submission errors here and propagate into subsequent ioend 1373598ecfbaSChristoph Hellwig * submissions. 1374598ecfbaSChristoph Hellwig */ 1375598ecfbaSChristoph Hellwig list_for_each_entry_safe(ioend, next, &submit_list, io_list) { 1376598ecfbaSChristoph Hellwig int error2; 1377598ecfbaSChristoph Hellwig 1378598ecfbaSChristoph Hellwig list_del_init(&ioend->io_list); 1379598ecfbaSChristoph Hellwig error2 = iomap_submit_ioend(wpc, ioend, error); 1380598ecfbaSChristoph Hellwig if (error2 && !error) 1381598ecfbaSChristoph Hellwig error = error2; 1382598ecfbaSChristoph Hellwig } 1383598ecfbaSChristoph Hellwig 1384598ecfbaSChristoph Hellwig /* 1385598ecfbaSChristoph Hellwig * We can end up here with no error and nothing to write only if we race 1386598ecfbaSChristoph Hellwig * with a partial page truncate on a sub-page block sized filesystem. 1387598ecfbaSChristoph Hellwig */ 1388598ecfbaSChristoph Hellwig if (!count) 1389598ecfbaSChristoph Hellwig end_page_writeback(page); 1390598ecfbaSChristoph Hellwig done: 1391598ecfbaSChristoph Hellwig mapping_set_error(page->mapping, error); 1392598ecfbaSChristoph Hellwig return error; 1393598ecfbaSChristoph Hellwig } 1394598ecfbaSChristoph Hellwig 1395598ecfbaSChristoph Hellwig /* 1396598ecfbaSChristoph Hellwig * Write out a dirty page. 1397598ecfbaSChristoph Hellwig * 1398f1f264b4SAndreas Gruenbacher * For delalloc space on the page, we need to allocate space and flush it. 1399f1f264b4SAndreas Gruenbacher * For unwritten space on the page, we need to start the conversion to 1400598ecfbaSChristoph Hellwig * regular allocated space. 1401598ecfbaSChristoph Hellwig */ 1402598ecfbaSChristoph Hellwig static int 1403598ecfbaSChristoph Hellwig iomap_do_writepage(struct page *page, struct writeback_control *wbc, void *data) 1404598ecfbaSChristoph Hellwig { 1405598ecfbaSChristoph Hellwig struct iomap_writepage_ctx *wpc = data; 1406598ecfbaSChristoph Hellwig struct inode *inode = page->mapping->host; 1407598ecfbaSChristoph Hellwig pgoff_t end_index; 1408598ecfbaSChristoph Hellwig u64 end_offset; 1409598ecfbaSChristoph Hellwig loff_t offset; 1410598ecfbaSChristoph Hellwig 14111ac99452SMatthew Wilcox (Oracle) trace_iomap_writepage(inode, page_offset(page), PAGE_SIZE); 1412598ecfbaSChristoph Hellwig 1413598ecfbaSChristoph Hellwig /* 1414f1f264b4SAndreas Gruenbacher * Refuse to write the page out if we're called from reclaim context. 1415598ecfbaSChristoph Hellwig * 1416598ecfbaSChristoph Hellwig * This avoids stack overflows when called from deeply used stacks in 1417598ecfbaSChristoph Hellwig * random callers for direct reclaim or memcg reclaim. We explicitly 1418598ecfbaSChristoph Hellwig * allow reclaim from kswapd as the stack usage there is relatively low. 1419598ecfbaSChristoph Hellwig * 1420598ecfbaSChristoph Hellwig * This should never happen except in the case of a VM regression so 1421598ecfbaSChristoph Hellwig * warn about it. 1422598ecfbaSChristoph Hellwig */ 1423598ecfbaSChristoph Hellwig if (WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) == 1424598ecfbaSChristoph Hellwig PF_MEMALLOC)) 1425598ecfbaSChristoph Hellwig goto redirty; 1426598ecfbaSChristoph Hellwig 1427598ecfbaSChristoph Hellwig /* 1428598ecfbaSChristoph Hellwig * Is this page beyond the end of the file? 1429598ecfbaSChristoph Hellwig * 1430598ecfbaSChristoph Hellwig * The page index is less than the end_index, adjust the end_offset 1431598ecfbaSChristoph Hellwig * to the highest offset that this page should represent. 1432598ecfbaSChristoph Hellwig * ----------------------------------------------------- 1433598ecfbaSChristoph Hellwig * | file mapping | <EOF> | 1434598ecfbaSChristoph Hellwig * ----------------------------------------------------- 1435598ecfbaSChristoph Hellwig * | Page ... | Page N-2 | Page N-1 | Page N | | 1436598ecfbaSChristoph Hellwig * ^--------------------------------^----------|-------- 1437598ecfbaSChristoph Hellwig * | desired writeback range | see else | 1438598ecfbaSChristoph Hellwig * ---------------------------------^------------------| 1439598ecfbaSChristoph Hellwig */ 1440598ecfbaSChristoph Hellwig offset = i_size_read(inode); 1441598ecfbaSChristoph Hellwig end_index = offset >> PAGE_SHIFT; 1442598ecfbaSChristoph Hellwig if (page->index < end_index) 1443598ecfbaSChristoph Hellwig end_offset = (loff_t)(page->index + 1) << PAGE_SHIFT; 1444598ecfbaSChristoph Hellwig else { 1445598ecfbaSChristoph Hellwig /* 1446598ecfbaSChristoph Hellwig * Check whether the page to write out is beyond or straddles 1447598ecfbaSChristoph Hellwig * i_size or not. 1448598ecfbaSChristoph Hellwig * ------------------------------------------------------- 1449598ecfbaSChristoph Hellwig * | file mapping | <EOF> | 1450598ecfbaSChristoph Hellwig * ------------------------------------------------------- 1451598ecfbaSChristoph Hellwig * | Page ... | Page N-2 | Page N-1 | Page N | Beyond | 1452598ecfbaSChristoph Hellwig * ^--------------------------------^-----------|--------- 1453598ecfbaSChristoph Hellwig * | | Straddles | 1454598ecfbaSChristoph Hellwig * ---------------------------------^-----------|--------| 1455598ecfbaSChristoph Hellwig */ 1456598ecfbaSChristoph Hellwig unsigned offset_into_page = offset & (PAGE_SIZE - 1); 1457598ecfbaSChristoph Hellwig 1458598ecfbaSChristoph Hellwig /* 1459f1f264b4SAndreas Gruenbacher * Skip the page if it's fully outside i_size, e.g. due to a 1460f1f264b4SAndreas Gruenbacher * truncate operation that's in progress. We must redirty the 1461598ecfbaSChristoph Hellwig * page so that reclaim stops reclaiming it. Otherwise 1462598ecfbaSChristoph Hellwig * iomap_vm_releasepage() is called on it and gets confused. 1463598ecfbaSChristoph Hellwig * 1464f1f264b4SAndreas Gruenbacher * Note that the end_index is unsigned long. If the given 1465f1f264b4SAndreas Gruenbacher * offset is greater than 16TB on a 32-bit system then if we 1466f1f264b4SAndreas Gruenbacher * checked if the page is fully outside i_size with 1467f1f264b4SAndreas Gruenbacher * "if (page->index >= end_index + 1)", "end_index + 1" would 1468f1f264b4SAndreas Gruenbacher * overflow and evaluate to 0. Hence this page would be 1469f1f264b4SAndreas Gruenbacher * redirtied and written out repeatedly, which would result in 1470f1f264b4SAndreas Gruenbacher * an infinite loop; the user program performing this operation 1471f1f264b4SAndreas Gruenbacher * would hang. Instead, we can detect this situation by 1472f1f264b4SAndreas Gruenbacher * checking if the page is totally beyond i_size or if its 1473598ecfbaSChristoph Hellwig * offset is just equal to the EOF. 1474598ecfbaSChristoph Hellwig */ 1475598ecfbaSChristoph Hellwig if (page->index > end_index || 1476598ecfbaSChristoph Hellwig (page->index == end_index && offset_into_page == 0)) 1477598ecfbaSChristoph Hellwig goto redirty; 1478598ecfbaSChristoph Hellwig 1479598ecfbaSChristoph Hellwig /* 1480598ecfbaSChristoph Hellwig * The page straddles i_size. It must be zeroed out on each 1481598ecfbaSChristoph Hellwig * and every writepage invocation because it may be mmapped. 1482598ecfbaSChristoph Hellwig * "A file is mapped in multiples of the page size. For a file 1483598ecfbaSChristoph Hellwig * that is not a multiple of the page size, the remaining 1484598ecfbaSChristoph Hellwig * memory is zeroed when mapped, and writes to that region are 1485598ecfbaSChristoph Hellwig * not written out to the file." 1486598ecfbaSChristoph Hellwig */ 1487598ecfbaSChristoph Hellwig zero_user_segment(page, offset_into_page, PAGE_SIZE); 1488598ecfbaSChristoph Hellwig 1489598ecfbaSChristoph Hellwig /* Adjust the end_offset to the end of file */ 1490598ecfbaSChristoph Hellwig end_offset = offset; 1491598ecfbaSChristoph Hellwig } 1492598ecfbaSChristoph Hellwig 1493598ecfbaSChristoph Hellwig return iomap_writepage_map(wpc, wbc, inode, page, end_offset); 1494598ecfbaSChristoph Hellwig 1495598ecfbaSChristoph Hellwig redirty: 1496598ecfbaSChristoph Hellwig redirty_page_for_writepage(wbc, page); 1497598ecfbaSChristoph Hellwig unlock_page(page); 1498598ecfbaSChristoph Hellwig return 0; 1499598ecfbaSChristoph Hellwig } 1500598ecfbaSChristoph Hellwig 1501598ecfbaSChristoph Hellwig int 1502598ecfbaSChristoph Hellwig iomap_writepage(struct page *page, struct writeback_control *wbc, 1503598ecfbaSChristoph Hellwig struct iomap_writepage_ctx *wpc, 1504598ecfbaSChristoph Hellwig const struct iomap_writeback_ops *ops) 1505598ecfbaSChristoph Hellwig { 1506598ecfbaSChristoph Hellwig int ret; 1507598ecfbaSChristoph Hellwig 1508598ecfbaSChristoph Hellwig wpc->ops = ops; 1509598ecfbaSChristoph Hellwig ret = iomap_do_writepage(page, wbc, wpc); 1510598ecfbaSChristoph Hellwig if (!wpc->ioend) 1511598ecfbaSChristoph Hellwig return ret; 1512598ecfbaSChristoph Hellwig return iomap_submit_ioend(wpc, wpc->ioend, ret); 1513598ecfbaSChristoph Hellwig } 1514598ecfbaSChristoph Hellwig EXPORT_SYMBOL_GPL(iomap_writepage); 1515598ecfbaSChristoph Hellwig 1516598ecfbaSChristoph Hellwig int 1517598ecfbaSChristoph Hellwig iomap_writepages(struct address_space *mapping, struct writeback_control *wbc, 1518598ecfbaSChristoph Hellwig struct iomap_writepage_ctx *wpc, 1519598ecfbaSChristoph Hellwig const struct iomap_writeback_ops *ops) 1520598ecfbaSChristoph Hellwig { 1521598ecfbaSChristoph Hellwig int ret; 1522598ecfbaSChristoph Hellwig 1523598ecfbaSChristoph Hellwig wpc->ops = ops; 1524598ecfbaSChristoph Hellwig ret = write_cache_pages(mapping, wbc, iomap_do_writepage, wpc); 1525598ecfbaSChristoph Hellwig if (!wpc->ioend) 1526598ecfbaSChristoph Hellwig return ret; 1527598ecfbaSChristoph Hellwig return iomap_submit_ioend(wpc, wpc->ioend, ret); 1528598ecfbaSChristoph Hellwig } 1529598ecfbaSChristoph Hellwig EXPORT_SYMBOL_GPL(iomap_writepages); 1530598ecfbaSChristoph Hellwig 1531598ecfbaSChristoph Hellwig static int __init iomap_init(void) 1532598ecfbaSChristoph Hellwig { 1533598ecfbaSChristoph Hellwig return bioset_init(&iomap_ioend_bioset, 4 * (PAGE_SIZE / SECTOR_SIZE), 1534598ecfbaSChristoph Hellwig offsetof(struct iomap_ioend, io_inline_bio), 1535598ecfbaSChristoph Hellwig BIOSET_NEED_BVECS); 1536598ecfbaSChristoph Hellwig } 1537598ecfbaSChristoph Hellwig fs_initcall(iomap_init); 1538